Original saved answers, research setups and evaluation details. Open a question or a model to examine the results for yourself.
Claude Fable 5.1 answered 9 questions about NVDA with no browsing or supplied documents, and with Finbly research and no browsing. These are saved answers, not a live demo.
Scoring proof-v2 · evaluated 9 Sept 2026. Checklist and number-occurrence counts do not establish accuracy. Web-enabled answers may choose not to search; recorded search counts appear beside each answer.
| Across 9 questions | Same model, no research | Same model with Finbly research |
|---|---|---|
| Checklist matchescontextual text matches; inspect the supporting passages · 25 and 71 of 84 | 30% | 85% |
| Numbers in saved evidencescalar occurrences only; not claim verification · of 296 and 441 checked | 104 | 371 |
| Claim accuracynot assessed by the numeric occurrence checker | Not assessed | Not assessed |
| Numbers without a scalar matchno scalar match in the saved packet; not necessarily wrong | 192 | 70 |
| Research reference tagscategory labels in the answer; repeats counted | 0 | 231 |
| Input tokens per questionwhat the model read | 130 | 11.0k |
| Cost per questionmodel list prices; a dash means no published rate | $0.14 | $0.22 |
Saved model outputs are preserved. Underlines describe automated numeric checks; evidence tags identify research categories and are not direct source citations.
Short answer: yes — it's held in the low-to-mid 70s and is drifting back up, after the one-off H20 hit.
The trajectory (non-GAAP GM)
So the true operating range has been ~71–74% through the Blackwell transition, with management guiding back to the mid-70s and saying they expect to hold that into FY27.
Why it dipped
Why it's recovering
What to watch
Bottom line: Gross margin is holding — ~73–75% is the new normal, down from the 78% Hopper peak but stable, with the Q1 FY26 collapse being an accounting charge rather than an economic one.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Bottom line: Gross margin is holding – ~73–75% is the new normal, down from the 78% Hopper peak but stable, with the Q1 FY26 collapse being an accounting charge rather than an economic one.
Why it dipped - Blackwell (GB200/GB300) is a full-rack system with much higher BOM content (CPUs, NVLink switches, liquid cooling, memory) – more pass-through cost than a Hopper HGX board. - Early yield/rework on the initial GB200 rack versions. - Mix shift toward lower-margin Networking and system-level revenue.
Yes — for now. Gross margin is not just holding, it's at a cycle high, but management has already told you it won't stay there.
NVIDIA's reported GAAP gross margin for Q2 FY2027 (quarter ended 2026-07-26) was [state]">75.0% on $96.2B of revenue [Q2 FY2027]. That's up a hair from [Q1 FY2027]">74.9% in Q1 FY2027 (quarter ended 2026-04-26) [Q1 FY2027] and up ~260 basis points from [Q2 FY2026]">72.4% in the year-ago Q2 FY2026 (quarter ended 2025-07-27) [Q2 FY2026] — my arithmetic on those two reported figures. For reference, the full FY2026 figure of [state]">71.1% looks weaker only because of a one-time ~$4.5B H20/China write-down that dragged Q1 FY2026 down to [Q1 FY2026]">60.5% [F, quarterly/NVDA]; ex-charge it ran ~73% [F]. So the trend into the latest print was clean expansion, with the mid-70s intact.
The problem is the guidance, not the print. On the Aug-26 call management guided Q3 FY2027 GAAP and non-GAAP gross margin to [guidance]">74.0% ± 50bp [guidance], then said margins will bottom at 71–72% in Q4 FY2027 and recover only to 72–73% for FY2028 [earnings recap agent]. From the reported [state]">75.0% to the guided 71–72% trough, that's a 300–400bp step down. The stated cause is memory cost inflation: the CFO said memory price increases exceeded prior expectations and are headed even higher [earnings recap agent]. This lands on NVIDIA's own cost of goods — the company is fabless, but it buys HBM memory as a component, so a DRAM price spike is a direct input-cost hit rather than something absorbed upstream. Management framed it as manageable and cyclical (the deep-dive calls it "structural for as long as the DRAM cycle lasts" [F]), but it is a real reset, not a China artifact.
Two balance-sheet facts frame how the margin should be read. First, inventories rose to [Q2 FY2027]">$31.6B in Q2 from [Q1 FY2027]">$25.8B in Q1 and [Q2 FY2026]">$15.0B a year ago [Q2 FY2027] [Q1 FY2027] [Q2 FY2026] — consistent with building Vera Rubin, which has begun production shipments with orders from all major hyperscalers [earnings recap agent], but also meaning higher-cost memory is now sitting on the books ahead of the guided trough. Second, receivables jumped to [Q2 FY2027]">$63.1B with DSO stretching to ~60 days from ~45 days in Q1 [Q2 FY2027] — not a margin item, but a sign the quality of the revenue behind that margin is worth monitoring.
On the watch list, the "gm_slip" trigger sits at a 73% threshold and is currently OK at [state]">75.0% — but it is scheduled to fire on management's own Q4 FY2027 guidance of 71–72% [watch]. Meanwhile the stock's reverse-DCF-implied FCF growth is a modeled [state]">29.9% a year for a decade [reverse-DCF], so the market is already pricing very high growth; a margin trough that lingers eats into that cushion. The stock still rose [earnings recap agent]">8.7% on the print despite the margin reset [earnings recap agent], so investors so far accept "volume growth trumps a few points of margin."
What to watch: (1) Q3 FY2027 GAAP gross margin on Nov-18 vs the [guidance]">74.0% ± 50bp guide — a miss below 73.5% says memory is worse than modeled [guidance]; (2) whether the Q4 trough language stays at 71–72% or drifts lower, and whether the FY2028 72–73% recovery is reiterated [earnings recap agent]; (3) inventory vs revenue — if inventories keep growing faster than sales beyond [Q2 FY2027]">$31.6B [Q2 FY2027], it raises the risk of high-cost stock being written down.
Bottom line: The margin held at a record [state]">75.0% in the latest quarter, but by management's own guidance it heads to a 71–72% trough within two quarters on memory costs — so "holding" is true today and already scheduled to be false.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
NVIDIA's reported GAAP gross margin for Q2 FY2027 (quarter ended 2026-07-26) was 75% on $96.2B of revenue Q2 FY2027. That's up a hair from 74.9% in Q1 FY2027 (quarter ended 2026-04-26) Q1 FY2027 and up ~260 basis points from 72.4% in the year-ago Q2 FY2026 (quarter ended 2025-07-27) Q2 FY2026 – my arithmetic on those two reported figures. For reference, the full FY2026 figure of 71.1% looks weaker only because of a one-time ~$4.5B H20/China write-down that dragged Q1 FY2026 down to 60.5% ; ex-charge it ran ~73% . So the trend into the latest print was clean expansion, with the mid-70s intact.
NVIDIA's reported GAAP gross margin for Q2 FY2027 (quarter ended 2026-07-26) was 75% on $96.2B of revenue Q2 FY2027. That's up a hair from 74.9% in Q1 FY2027 (quarter ended 2026-04-26) Q1 FY2027 and up ~260 basis points from 72.4% in the year-ago Q2 FY2026 (quarter ended 2025-07-27) Q2 FY2026 – my arithmetic on those two reported figures. For reference, the full FY2026 figure of 71.1% looks weaker only because of a one-time ~$4.5B H20/China write-down that dragged Q1 FY2026 down to 60.5% ; ex-charge it ran ~73% . So the trend into the latest print was clean expansion, with the mid-70s intact.
NVIDIA's reported GAAP gross margin for Q2 FY2027 (quarter ended 2026-07-26) was 75% on $96.2B of revenue Q2 FY2027. That's up a hair from 74.9% in Q1 FY2027 (quarter ended 2026-04-26) Q1 FY2027 and up ~260 basis points from 72.4% in the year-ago Q2 FY2026 (quarter ended 2025-07-27) Q2 FY2026 – my arithmetic on those two reported figures. For reference, the full FY2026 figure of 71.1% looks weaker only because of a one-time ~$4.5B H20/China write-down that dragged Q1 FY2026 down to 60.5% ; ex-charge it ran ~73% . So the trend into the latest print was clean expansion, with the mid-70s intact.
The problem is the guidance, not the print. On the Aug-26 call management guided Q3 FY2027 GAAP and non-GAAP gross margin to 74% ± 50bp , then said margins will bottom at 71–72% in Q4 FY2027 and recover only to 72–73% for FY2028 . From the reported 75% to the guided 71–72% trough, that's a 300–400bp step down. The stated cause is memory cost inflation: the CFO said memory price increases exceeded prior expectations and are headed even higher . This lands on NVIDIA's own cost of goods – the company is fabless, but it buys HBM memory as a component, so a DRAM price spike is a direct input-cost hit rather than something absorbed upstream. Management framed it as manageable and cyclical (the deep-dive calls it "structural for as long as the DRAM cycle lasts" ), but it is a real reset, not a China artifact.
The problem is the guidance, not the print. On the Aug-26 call management guided Q3 FY2027 GAAP and non-GAAP gross margin to 74% ± 50bp , then said margins will bottom at 71–72% in Q4 FY2027 and recover only to 72–73% for FY2028 . From the reported 75% to the guided 71–72% trough, that's a 300–400bp step down. The stated cause is memory cost inflation: the CFO said memory price increases exceeded prior expectations and are headed even higher . This lands on NVIDIA's own cost of goods – the company is fabless, but it buys HBM memory as a component, so a DRAM price spike is a direct input-cost hit rather than something absorbed upstream. Management framed it as manageable and cyclical (the deep-dive calls it "structural for as long as the DRAM cycle lasts" ), but it is a real reset, not a China artifact.
The problem is the guidance, not the print. On the Aug-26 call management guided Q3 FY2027 GAAP and non-GAAP gross margin to 74% ± 50bp , then said margins will bottom at 71–72% in Q4 FY2027 and recover only to 72–73% for FY2028 . From the reported 75% to the guided 71–72% trough, that's a 300–400bp step down. The stated cause is memory cost inflation: the CFO said memory price increases exceeded prior expectations and are headed even higher . This lands on NVIDIA's own cost of goods – the company is fabless, but it buys HBM memory as a component, so a DRAM price spike is a direct input-cost hit rather than something absorbed upstream. Management framed it as manageable and cyclical (the deep-dive calls it "structural for as long as the DRAM cycle lasts" ), but it is a real reset, not a China artifact.
The problem is the guidance, not the print. On the Aug-26 call management guided Q3 FY2027 GAAP and non-GAAP gross margin to 74% ± 50bp , then said margins will bottom at 71–72% in Q4 FY2027 and recover only to 72–73% for FY2028 . From the reported 75% to the guided 71–72% trough, that's a 300–400bp step down. The stated cause is memory cost inflation: the CFO said memory price increases exceeded prior expectations and are headed even higher . This lands on NVIDIA's own cost of goods – the company is fabless, but it buys HBM memory as a component, so a DRAM price spike is a direct input-cost hit rather than something absorbed upstream. Management framed it as manageable and cyclical (the deep-dive calls it "structural for as long as the DRAM cycle lasts" ), but it is a real reset, not a China artifact.
On the watch list, the "gmslip" trigger sits at a 73% threshold and is currently OK at 75% – but it is scheduled to fire on management's own Q4 FY2027 guidance of 71–72% . Meanwhile the stock's reverse-DCF-implied FCF growth is a modeled 29.9% a year for a decade , so the market is already pricing very high growth; a margin trough that lingers eats into that cushion. The stock still rose 8.7% on the print despite the margin reset , so investors so far accept "volume growth trumps a few points of margin."
Two balance-sheet facts frame how the margin should be read. First, inventories rose to $31.6B in Q2 from $25.8B in Q1 and $15.0B a year ago Q2 FY2027 Q1 FY2027 Q2 FY2026 – consistent with building Vera Rubin, which has begun production shipments with orders from all major hyperscalers , but also meaning higher-cost memory is now sitting on the books ahead of the guided trough. Second, receivables jumped to $63.1B with DSO stretching to ~60 days from ~45 days in Q1 Q2 FY2027 – not a margin item, but a sign the quality of the revenue behind that margin is worth monitoring.
Short answer (as of my last data, roughly mid-2025 — verify live, NVDA numbers move fast): consensus 12-month target ~$175–$200, and the stock trades ~30x forward (next-12-month) EPS, falling to the low-20s on FY2027 numbers.
Price targets
Forward P/E
Estimate trajectory
What's baked in / risks the Street flags
Bottom line: Street is near-unanimously bullish with targets ~15% above the stock, and at ~30x forward/~24x FY27 the multiple is at the low end of NVDA's own history — the stock isn't expensive on Street numbers; the risk is the Street numbers.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Price targets - Coverage: ~60+ analysts; ~90% Buy/Overweight, a handful of Holds, essentially no Sells. - Consensus target: ~$175–$200 range, implying low-to-mid teens percent upside depending on where the stock sits. - Street high: ~$220–$250 (Rosenblatt, Loop-type bulls). Street low: ~$100–$135 (DA Davidson/Seaport-type Neutrals). - Targets have consistently trailed the stock – the pattern is the stock runs, then targets get marked up after earnings.
Analysts see ~[state]">50% upside and price NVDA at roughly [state]">25× this year's earnings. The average price target is [estimate]">$345.21 (consensus, 22 analysts), median [estimate]">$322.50, range $270–$515 [estimate] — the average works out to [state]">49.9% above the [state]">$230.36 price (as of 2026-09-04) [state]. Notably, even the lowest target on the Street ($270) sits above where the stock trades today.
Forward P/E: [state]">24.9×, built as the [state]">$230.36 price divided by consensus FY2027 EPS of [estimate]">$9.26 (30 analysts; FY2027 ends January 2027) [estimate]. That compares with a trailing P/E of [state]">29.1× (TTM through Q2 FY2027, quarter ended July 26, 2026) [state]. Older research put the forward multiple at ~21–24× when the stock was ~$208–212; the [state]">24.9× figure is the current one. Either way, it is far below the stock's own ~[F]), NVDA trades at a **~21–24× forward P/E [F]** versus its own **~63× five-year average — its own P/E has ranged ~38–120× since FY2022 [finbly, verified]** — and is up only ~high-single-digits YTD while **AMD is +132% YTD** (finbly market data). The">63× five-year average P/E (range ~38–120× since FY2022) [F] — the market has let earnings grow into the price rather than re-rate it.
The out-years are where the expectations really sit. Consensus EPS runs [estimate]">$9.26 (FY2027) → [estimate]">$15.64 (FY2028, 32 analysts) → [estimate]">$21.16 (FY2029, 19 analysts) → [estimate]">$23 (FY2030, 10 analysts) [estimate]. Deriving from those inputs: at [state]">$230.36 the stock is ~14.7× FY2028 EPS and ~10.9× FY2029 EPS — multiples normally attached to no-growth companies, not one whose FY2028 EPS is expected to rise ~69% ([estimate]">$15.64 vs [estimate]">$9.26). Consensus revenue tells the same story: $408.4B FY2027 → $685.9B FY2028 → $904.4B FY2029 [estimate], which lines up with management's own call comment that FY2028 revenue grows ~[V, call].">70%, described as supply-constrained [V, call].
Direction of revisions is up. Between the Aug-31 and Sep-7 snapshots, five estimate changes were detected, all upward, zero downward: FY2028 EPS [estimate]">+0.7%, FY2029 EPS [estimate]">+1.7%, FY2029 revenue [estimate]">+1.7% [estimate]. Ratings are 2 strong-buy / 58 buy / 16 hold / 3 sell [estimate]. Next print is November 18, 2026, with consensus EPS of [estimate]">$2.47 [estimate], essentially flat versus the $2.46 diluted GAAP EPS reported for Q2 FY2027 [F, quarterly/NVDA].
Two things frame those numbers. First, margins: Q3 FY2027 gross margin is guided to [guidance]">74.0% ±50bp on [guidance]">$108.0B ±[guidance]">2% of revenue [guidance], but management guided Q4 FY2027 to a 71–72% trough and FY2028 to 72–73% on memory costs [F, Aug-26 guidance] — so out-year EPS estimates carry a margin headwind that consensus must already absorb. Second, earnings quality: in Q1 FY2027 roughly $16B of unrealized gains on equity stakes in customers flowed into GAAP net income, making GAAP EPS ~$12.8B higher than non-GAAP [F], so the trailing [state]">29.1× rests partly on marks, not chips. And the price still embeds a lot: the reverse DCF requires [state]">29.9% annual FCF growth for a decade (modeled) [reverse-DCF], while the low end of the target range is still 17% above spot (270/[estimate]">230.36).
What to watch: (1) Whether the Nov-18 print lands at or above the [guidance]">$108.0B guide and [estimate]">$2.47 EPS consensus; (2) any downward revisions to FY2028 EPS as the 71–72% Q4 margin trough gets modeled; (3) the hyperscaler 2027 capex guidance cycle — a coordinated cut is the single trigger that would break the $685.9B FY2028 revenue consensus [watch].
Bottom line: The Street expects ~[estimate]">$345 (≈[state]">+50%) and values NVDA at [state]">24.9× FY2027 and ~15× FY2028 earnings — cheap on its own history, with estimates still rising — but the multiple is only cheap if the guided margin trough and the debt-financed customer buildout don't derail the ~[V, call].">70% FY2028 growth everyone is penciling in.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Analysts see ~50% upside and price NVDA at roughly 25× this year's earnings. The average price target is $345.21 (consensus, 22 analysts), median $322.50, range $270–$515 – the average works out to 49.9% above the $230.36 price (as of 2026-09-04) . Notably, even the lowest target on the Street ($270) sits above where the stock trades today.
Analysts see ~50% upside and price NVDA at roughly 25× this year's earnings. The average price target is $345.21 (consensus, 22 analysts), median $322.50, range $270–$515 – the average works out to 49.9% above the $230.36 price (as of 2026-09-04) . Notably, even the lowest target on the Street ($270) sits above where the stock trades today.
Analysts see ~50% upside and price NVDA at roughly 25× this year's earnings. The average price target is $345.21 (consensus, 22 analysts), median $322.50, range $270–$515 – the average works out to 49.9% above the $230.36 price (as of 2026-09-04) . Notably, even the lowest target on the Street ($270) sits above where the stock trades today.
Forward P/E: 24.9×, built as the $230.36 price divided by consensus FY2027 EPS of $9.26 (30 analysts; FY2027 ends January 2027) . That compares with a trailing P/E of 29.1× (TTM through Q2 FY2027, quarter ended July 26, 2026) . Older research put the forward multiple at ~21–24× when the stock was ~$208–212; the 24.9× figure is the current one. Either way, it is far below the stock's own ~63× five-year average P/E (range ~38–120× since FY2022) – the market has let earnings grow into the price rather than re-rate it.
Forward P/E: 24.9×, built as the $230.36 price divided by consensus FY2027 EPS of $9.26 (30 analysts; FY2027 ends January 2027) . That compares with a trailing P/E of 29.1× (TTM through Q2 FY2027, quarter ended July 26, 2026) . Older research put the forward multiple at ~21–24× when the stock was ~$208–212; the 24.9× figure is the current one. Either way, it is far below the stock's own ~63× five-year average P/E (range ~38–120× since FY2022) – the market has let earnings grow into the price rather than re-rate it.
The out-years are where the expectations really sit. Consensus EPS runs $9.26 (FY2027) → $15.64 (FY2028, 32 analysts) → $21.16 (FY2029, 19 analysts) → $23 (FY2030, 10 analysts) . Deriving from those inputs: at $230.36 the stock is ~14.7× FY2028 EPS and ~10.9× FY2029 EPS – multiples normally attached to no-growth companies, not one whose FY2028 EPS is expected to rise ~69% ($15.64 vs $9.26). Consensus revenue tells the same story: $408.4B FY2027 → $685.9B FY2028 → $904.4B FY2029 , which lines up with management's own call comment that FY2028 revenue grows ~70%, described as supply-constrained .
Analysts see ~50% upside and price NVDA at roughly 25× this year's earnings. The average price target is $345.21 (consensus, 22 analysts), median $322.50, range $270–$515 – the average works out to 49.9% above the $230.36 price (as of 2026-09-04) . Notably, even the lowest target on the Street ($270) sits above where the stock trades today.
Direction of revisions is up. Between the Aug-31 and Sep-7 snapshots, five estimate changes were detected, all upward, zero downward: FY2028 EPS +0.7%, FY2029 EPS +1.7%, FY2029 revenue +1.7% . Ratings are 2 strong-buy / 58 buy / 16 hold / 3 sell . Next print is November 18, 2026, with consensus EPS of $2.47 , essentially flat versus the $2.46 diluted GAAP EPS reported for Q2 FY2027 .
Direction of revisions is up. Between the Aug-31 and Sep-7 snapshots, five estimate changes were detected, all upward, zero downward: FY2028 EPS +0.7%, FY2029 EPS +1.7%, FY2029 revenue +1.7% . Ratings are 2 strong-buy / 58 buy / 16 hold / 3 sell . Next print is November 18, 2026, with consensus EPS of $2.47 , essentially flat versus the $2.46 diluted GAAP EPS reported for Q2 FY2027 .
Most recent quarter I can confirm: Q3 FY2026 (quarter ended Oct 26, 2025; reported Nov 19, 2025). Revenue was $57.0B, +62% y/y and +22% q/q, a beat against the $54B guide and ~$55B consensus. If Q4 FY26 (due ~late Feb 2026) has already printed, check that release — guidance for it was $65B.
Headline numbers
Segments
Guidance (Q4 FY26)
What mattered
Bottom line: A clean beat-and-raise — $57B revenue, $51B data center, 73.6% margins, $65B guide — with the incremental $500B backlog disclosure doing more for the multi-year story than the quarter itself.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Guidance (Q4 FY26) - Revenue $65.0B ±2% → implies ~+14% q/q, ~+65% y/y - Non-GAAP gross margin ~75% - Non-GAAP opex ~$5.0B
Q2 FY2027 (quarter ended July 26, 2026) was a blowout on the top line — reported GAAP revenue of $96.2B, up 106% from a year ago and about [guidance]">18% from Q1's $81.6B (96.2 ÷ 81.6), with GAAP gross margin at [state]">75.0% and net income of $59.7B [Q2 FY2027]. The catch is what came with it: management guided gross margin down from here.
The print vs. the bar. Non-GAAP EPS of [F, earnings recap]: non-GAAP EPS $2.22 vs $2.09; revenue $96.2B vs $92.3B; shares +8.7% next">$2.22 beat the [F, earnings recap]: non-GAAP EPS $2.22 vs $2.09; revenue $96.2B vs $92.3B; shares +8.7% next">$2.09 consensus by 6.2%, and revenue of $96.22B beat the $92.27B consensus by 4.3% — the seventh straight quarter beating on both lines [earnings recap agent]. The stock rose [earnings recap agent]">8.7% the next session on [earnings recap agent]">2.6x normal volume [earnings recap agent]. GAAP diluted EPS was $2.46 [F, quarterly/NVDA].
Trend. The five-quarter tape runs $44.1B → $46.7B → $57.0B → $81.6B → $96.2B, so growth is re-accelerating, not fading [Q1 FY2026] [Q2 FY2027]. Gross margin has climbed from [Q2 FY2026]">72.4% in the year-ago quarter (Q2 FY2026) and [Q1 FY2027]">74.9% last quarter to [state]">75.0% now [Q2 FY2026] [Q1 FY2027] [Q2 FY2027]. Net income more than doubled from [Q2 FY2026]">$26.4B a year ago [Q2 FY2026].
The margin reset — the real news. Guided Q3 FY2027 (quarter ending October 2026) revenue is [guidance]">$108.0B ±[guidance]">2%, with GAAP and non-GAAP gross margin of [guidance]">74.0% ±50bp [guidance]. That revenue guide sat 0.6% below the $108.65B consensus, which is why the scorecard called it a "beat" rather than a "beat-and-raise" [earnings recap agent]. More important, management said gross margin will bottom at 71–72% in Q4 FY2027 and run 72–73% in FY2028, because memory (DRAM/HBM) price increases exceeded expectations and are heading higher — NVIDIA is fabless, so this is an input-cost squeeze on its own COGS via supplier pricing [earnings recap agent]. Offsetting that, management guided FY2028 revenue growth of roughly [earnings recap agent]">70%, explicitly supply-constrained versus an estimated ~100% true demand growth, and said the Vera Rubin platform has begun production shipments with orders from every major hyperscaler [earnings recap agent].
Balance sheet moves worth knowing. Receivables jumped to [Q2 FY2027]">$63.1B from [Q1 FY2027]">$40.7B, pushing days-sales-outstanding (how long customers take to pay) to ~60 days from ~45 [Q1 FY2027] [Q2 FY2027]. Inventories rose to [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B [Q2 FY2027]. Total debt quadrupled to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B after [Q2 FY2027]">$24.9B of issuance, even as cash rose to [Q2 FY2027]">$22.4B; year-to-date buybacks were [Q2 FY2027]">$39.0B and dividends [Q2 FY2027]">$6.3B [Q2 FY2027]. Management acknowledged nearly $50B of exposure to Frontier AI labs without investment-grade credit — the "circular financing" question of NVIDIA capital funding its own customers [earnings recap agent]. GAAP earnings also included $7.8B of Q2 gains on equity securities [F, quarterly/NVDA], so the reported net income is flattered by marks on those holdings.
What to watch: (1) Q3 gross margin vs. the [guidance]">74% guide and whether Q4 actually troughs at 71–72% — the gm_slip trigger is scheduled to fire on management's own outlook [watch]; (2) receivables/DSO at the November 18 print — another leg up from 60 days would sharpen the financing concern [estimate]; (3) any confirmation or denial of the reported ~$250B OpenAI deal structure [watch].
Bottom line: Demand is no longer the debate — a 106% y/y quarter and a [guidance]">$108B guide settle that — but the quarter shifted the fault lines to NVIDIA's own margin trough on memory costs and to how much of its customers' buying it is financing itself.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Q2 FY2027 (quarter ended July 26, 2026) was a blowout on the top line – reported GAAP revenue of $96.2B, up 106% from a year ago and about 18% from Q1's $81.6B (96.2 ÷ 81.6), with GAAP gross margin at 75% and net income of $59.7B Q2 FY2027. The catch is what came with it: management guided gross margin down from here.
Q2 FY2027 (quarter ended July 26, 2026) was a blowout on the top line – reported GAAP revenue of $96.2B, up 106% from a year ago and about 18% from Q1's $81.6B (96.2 ÷ 81.6), with GAAP gross margin at 75% and net income of $59.7B Q2 FY2027. The catch is what came with it: management guided gross margin down from here.
Q2 FY2027 (quarter ended July 26, 2026) was a blowout on the top line – reported GAAP revenue of $96.2B, up 106% from a year ago and about 18% from Q1's $81.6B (96.2 ÷ 81.6), with GAAP gross margin at 75% and net income of $59.7B Q2 FY2027. The catch is what came with it: management guided gross margin down from here.
Q2 FY2027 (quarter ended July 26, 2026) was a blowout on the top line – reported GAAP revenue of $96.2B, up 106% from a year ago and about 18% from Q1's $81.6B (96.2 ÷ 81.6), with GAAP gross margin at 75% and net income of $59.7B Q2 FY2027. The catch is what came with it: management guided gross margin down from here.
The print vs. the bar. Non-GAAP EPS of $2.22 beat the $2.09 consensus by 6.2%, and revenue of $96.22B beat the $92.27B consensus by 4.3% – the seventh straight quarter beating on both lines . The stock rose 8.7% the next session on 2.6x normal volume . GAAP diluted EPS was $2.46 .
The margin reset – the real news. Guided Q3 FY2027 (quarter ending October 2026) revenue is $108.0B ±2%, with GAAP and non-GAAP gross margin of 74% ±50bp . That revenue guide sat 0.6% below the $108.65B consensus, which is why the scorecard called it a "beat" rather than a "beat-and-raise" . More important, management said gross margin will bottom at 71–72% in Q4 FY2027 and run 72–73% in FY2028, because memory (DRAM/HBM) price increases exceeded expectations and are heading higher – NVIDIA is fabless, so this is an input-cost squeeze on its own COGS via supplier pricing . Offsetting that, management guided FY2028 revenue growth of roughly 70%, explicitly supply-constrained versus an estimated ~100% true demand growth, and said the Vera Rubin platform has begun production shipments with orders from every major hyperscaler .
Balance sheet moves worth knowing. Receivables jumped to $63.1B from $40.7B, pushing days-sales-outstanding (how long customers take to pay) to ~60 days from ~45 Q1 FY2027 Q2 FY2027. Inventories rose to $31.6B from $25.8B Q2 FY2027. Total debt quadrupled to $33.4B from $8.5B after $24.9B of issuance, even as cash rose to $22.4B; year-to-date buybacks were $39.0B and dividends $6.3B Q2 FY2027. Management acknowledged nearly $50B of exposure to Frontier AI labs without investment-grade credit – the "circular financing" question of NVIDIA capital funding its own customers . GAAP earnings also included $7.8B of Q2 gains on equity securities , so the reported net income is flattered by marks on those holdings.
Balance sheet moves worth knowing. Receivables jumped to $63.1B from $40.7B, pushing days-sales-outstanding (how long customers take to pay) to ~60 days from ~45 Q1 FY2027 Q2 FY2027. Inventories rose to $31.6B from $25.8B Q2 FY2027. Total debt quadrupled to $33.4B from $8.5B after $24.9B of issuance, even as cash rose to $22.4B; year-to-date buybacks were $39.0B and dividends $6.3B Q2 FY2027. Management acknowledged nearly $50B of exposure to Frontier AI labs without investment-grade credit – the "circular financing" question of NVIDIA capital funding its own customers . GAAP earnings also included $7.8B of Q2 gains on equity securities , so the reported net income is flattered by marks on those holdings.
The margin reset – the real news. Guided Q3 FY2027 (quarter ending October 2026) revenue is $108.0B ±2%, with GAAP and non-GAAP gross margin of 74% ±50bp . That revenue guide sat 0.6% below the $108.65B consensus, which is why the scorecard called it a "beat" rather than a "beat-and-raise" . More important, management said gross margin will bottom at 71–72% in Q4 FY2027 and run 72–73% in FY2028, because memory (DRAM/HBM) price increases exceeded expectations and are heading higher – NVIDIA is fabless, so this is an input-cost squeeze on its own COGS via supplier pricing . Offsetting that, management guided FY2028 revenue growth of roughly 70%, explicitly supply-constrained versus an estimated ~100% true demand growth, and said the Vera Rubin platform has begun production shipments with orders from every major hyperscaler .
Short answer: expensive on sales and absolute size, not expensive on earnings — ~30x forward EPS, ~48x trailing, for a business still growing [state]">50%+. (Figures anchored to a ~$140 share price / ~$3.4T market cap as of my latest data; scale the multiples to today's quote.)
Headline multiples
Versus peers and history
Where it is demanding
The math that matters Assume FY27 EPS of ~$5.75–6.00. At [state]">25x (a normal high-quality semi multiple) that's ~$145–150 — roughly flat from $140 over two years despite ~[state]">30% earnings growth. At 35x it's ~$200–210. The return is almost entirely a multiple call, not an earnings call. Downside case: a capex digestion year (growth slows to 15–20%) plus compression to 20x on ~$5 EPS = ~$100.
Bottom line: NVDA isn't priced for perfection on earnings (~30x forward, PEG <1), but it is priced for durability — the risk isn't the multiple, it's whether $200B+ in AI capex holds through FY27, and any wobble there compresses the multiple faster than earnings can offset it.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Versus peers and history - S&P 500 ~22x forward; SOX ~25–28x; MSFT ~30x; AVGO ~35x. NVDA's forward P/E sits in the pack, not above it. - Against its own history the forward multiple is near the low end of its 5-year range (25–70x) – not because the stock is cheap, but because earnings have compounded faster than the price.
Versus peers and history - S&P 500 ~22x forward; SOX ~25–28x; MSFT ~30x; AVGO ~35x. NVDA's forward P/E sits in the pack, not above it. - Against its own history the forward multiple is near the low end of its 5-year range (25–70x) – not because the stock is cheap, but because earnings have compounded faster than the price.
Demanding on cash flow, undemanding on earnings — that split is the whole answer. At [state]">$230.36 (as of Sept 4) and a [state]">$5.6T market cap, the modeled reverse DCF says the price embeds [state]">29.9% annual free-cash-flow growth for ten years [state]. Yet the stock trades at a [state]">29.1x trailing P/E (GAAP, TTM through Q2 FY2027, the quarter ended July 26, 2026) and a [state]">24.9x forward P/E on consensus FY2027 EPS of [estimate]">$9.26 [state][estimate] — a fraction of its own history.
The absolute bar. The reverse DCF (a model that backs out what growth the price requires) needs ~[state]">30% compounded FCF growth for a decade at a ~13.9% discount rate; an older run at $220.78 showed 29.4% [valuation agent over deterministic models; scenarios, not a target]. The trigger for "market demanding even more" is [watch]">32%, so there is a bit of room but not much [watch]. Peer comparison is the harsh version: on peer-median price-to-sales the modeled downside is [state]">-69.5%, past the [watch]">-65% alarm — meaning if NVDA were ever valued as "just a chip company," the premium collapses [state][watch]. Price-to-sales is [state]">18.5x and EV/EBITDA [state]">27.9x [state].
The relative bar. Against its own record, this is cheap: forward P/E of ~21–24x versus a ~[F]), NVDA trades at a **~21–24× forward P/E [F]** versus its own **~63× five-year average — its own P/E has ranged ~38–120× since FY2022 [finbly, verified]** — and is up only ~high-single-digits YTD while **AMD is +132% YTD** (finbly market data). The">63x five-year average, with the trailing P/E having ranged ~38–120x since FY2022 [F]. The stock is roughly flat year-to-date while AMD is up 130%+ — the market is letting earnings grow into the price rather than re-rating it. Consensus sell-side sees [state]">49.9% upside to an average target of [estimate]">$345 (range $270–$515, 22 analysts), and revisions over the last week were 5 up, 0 down [estimate]. On consensus FY2028 EPS of [estimate]">$15.64, the forward multiple falls to ~14.7x ([estimate]">230.36 / [estimate]">15.64, derived) [estimate].
What the fundamentals say about clearing the bar. Reported Q2 FY2027 revenue was $96.2B, +106% y/y and [guidance]">+18% from Q1's $81.6B, at a [state]">75.0% GAAP gross margin [Q2 FY2027]. Guided Q3 revenue is [guidance]">$108.0B ±[guidance]">2% with [guidance]">74.0% gross margin [guidance]; management also guided a 71–72% gross-margin trough in Q4 FY2027 and ~[V, call].">+70% FY2028 revenue [V, call]. FY2026 ROIC was [state]">66.8% against a ~13.9% WACC [state]. Those numbers make a [state]">30% FCF CAGR plausible near-term — the question is years 5–10.
Two cautions that make the cheap P/E less cheap than it looks. First, GAAP net income now includes large unrealized equity marks on customers NVIDIA funds — ~$16B in Q1 FY2027, another $7.8B in Q2 [F] — so the [state]">29.1x trailing multiple is flattered; non-GAAP EPS was [F, earnings recap]: non-GAAP EPS $2.22 vs $2.09; revenue $96.2B vs $92.3B; shares +8.7% next">$2.22 versus GAAP $2.46 last quarter [F]. Second, the balance sheet is shifting: total debt rose to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B, receivables jumped to [Q2 FY2027]">$63.1B (DSO ~60 days from ~45), and inventories to [Q2 FY2027]">$31.6B [Q2 FY2027]. Growth is being financed, not just harvested.
What to watch: whether reported Q3 FCF tracks the ~[state]">30% requirement; whether Q4 gross margin lands at or above the 71–72% guide (the gm_slip trigger is scheduled to fire); and hyperscaler 2027 capex guidance — a coordinated cut is the one event that flips the thesis [watch].
Bottom line: the price requires a decade of ~[state]">30% FCF growth and a premium-to-peers that is already stretched, but on next year's earnings NVDA is near its cheapest in a decade — demanding in the long run, undemanding in the short run, and the gap closes only if the customers' financing holds.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Demanding on cash flow, undemanding on earnings – that split is the whole answer. At $230.36 (as of Sept 4) and a $5.6T market cap, the modeled reverse DCF says the price embeds 29.9% annual free-cash-flow growth for ten years . Yet the stock trades at a 29.1x trailing P/E (GAAP, TTM through Q2 FY2027, the quarter ended July 26, 2026) and a 24.9x forward P/E on consensus FY2027 EPS of $9.26 – a fraction of its own history.
Demanding on cash flow, undemanding on earnings – that split is the whole answer. At $230.36 (as of Sept 4) and a $5.6T market cap, the modeled reverse DCF says the price embeds 29.9% annual free-cash-flow growth for ten years . Yet the stock trades at a 29.1x trailing P/E (GAAP, TTM through Q2 FY2027, the quarter ended July 26, 2026) and a 24.9x forward P/E on consensus FY2027 EPS of $9.26 – a fraction of its own history.
Demanding on cash flow, undemanding on earnings – that split is the whole answer. At $230.36 (as of Sept 4) and a $5.6T market cap, the modeled reverse DCF says the price embeds 29.9% annual free-cash-flow growth for ten years . Yet the stock trades at a 29.1x trailing P/E (GAAP, TTM through Q2 FY2027, the quarter ended July 26, 2026) and a 24.9x forward P/E on consensus FY2027 EPS of $9.26 – a fraction of its own history.
The absolute bar. The reverse DCF (a model that backs out what growth the price requires) needs ~30% compounded FCF growth for a decade at a ~13.9% discount rate; an older run at $220.78 showed 29.4% . The trigger for "market demanding even more" is 32%, so there is a bit of room but not much . Peer comparison is the harsh version: on peer-median price-to-sales the modeled downside is -69.5%, past the -65% alarm – meaning if NVDA were ever valued as "just a chip company," the premium collapses . Price-to-sales is 18.5x and EV/EBITDA 27.9x .
The absolute bar. The reverse DCF (a model that backs out what growth the price requires) needs ~30% compounded FCF growth for a decade at a ~13.9% discount rate; an older run at $220.78 showed 29.4% . The trigger for "market demanding even more" is 32%, so there is a bit of room but not much . Peer comparison is the harsh version: on peer-median price-to-sales the modeled downside is -69.5%, past the -65% alarm – meaning if NVDA were ever valued as "just a chip company," the premium collapses . Price-to-sales is 18.5x and EV/EBITDA 27.9x .
The absolute bar. The reverse DCF (a model that backs out what growth the price requires) needs ~30% compounded FCF growth for a decade at a ~13.9% discount rate; an older run at $220.78 showed 29.4% . The trigger for "market demanding even more" is 32%, so there is a bit of room but not much . Peer comparison is the harsh version: on peer-median price-to-sales the modeled downside is -69.5%, past the -65% alarm – meaning if NVDA were ever valued as "just a chip company," the premium collapses . Price-to-sales is 18.5x and EV/EBITDA 27.9x .
Demanding on cash flow, undemanding on earnings – that split is the whole answer. At $230.36 (as of Sept 4) and a $5.6T market cap, the modeled reverse DCF says the price embeds 29.9% annual free-cash-flow growth for ten years . Yet the stock trades at a 29.1x trailing P/E (GAAP, TTM through Q2 FY2027, the quarter ended July 26, 2026) and a 24.9x forward P/E on consensus FY2027 EPS of $9.26 – a fraction of its own history.
The relative bar. Against its own record, this is cheap: forward P/E of ~21–24x versus a ~63x five-year average, with the trailing P/E having ranged ~38–120x since FY2022 . The stock is roughly flat year-to-date while AMD is up 130%+ – the market is letting earnings grow into the price rather than re-rating it. Consensus sell-side sees 49.9% upside to an average target of $345 (range $270–$515, 22 analysts), and revisions over the last week were 5 up, 0 down . On consensus FY2028 EPS of $15.64, the forward multiple falls to ~14.7x (230.36 / 15.64, derived) .
Short version: the five that matter most, in order — hyperscaler capex durability, China/export policy, custom silicon (TPU/Trainium/ASICs), circular financing of the demand base, and single-point supply chain exposure to TSMC/HBM. Valuation is a sixth, but it's a symptom of the others. Figures below are as of my latest data; verify against the most recent 10-Q.
1. Customer concentration / capex cycle (biggest, ~[state]">50%+ of the equity story)
2. China / export controls
3. Custom silicon and AMD
4. Circular financing / fragility of the demand base
5. Supply chain / Taiwan
6. Execution and margins
7. Valuation as an amplifier
Bottom line: The stock is a leveraged bet on hyperscaler capex continuing to compound; everything else — China, ASICs, circular financing, TSMC — is a channel through which that bet can break, and the first sign of trouble will show up in a sequential capex guide-down from Microsoft/Meta/Google or a Rubin delay, not in NVDA's own print.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Short version: the five that matter most, in order – hyperscaler capex durability, China/export policy, custom silicon (TPU/Trainium/ASICs), circular financing of the demand base, and single-point supply chain exposure to TSMC/HBM. Valuation is a sixth, but it's a symptom of the others. Figures below are as of my latest data; verify against the most recent 10-Q.
1. Customer concentration / capex cycle (biggest, ~50%+ of the equity story) - Two direct customers were ~23% and ~16% of revenue in the most recent quarter; the top four US hyperscalers plus Oracle are roughly half of data center revenue. - Hyperscaler capex is running ~$400B annualized and guided higher for 2026. NVDA needs that number to keep rising, not just hold. A flat capex year ≈ a flat-to-down NVDA revenue year given price/mix. - The unknown is AI ROI at the application layer. Enterprise monetization is still thin relative to spend; if CFOs see it as an arms race with no payback, orders get pushed rather than cancelled – but a push is enough to break the multiple.
5. Supply chain / Taiwan - Essentially all leading-edge silicon comes from TSMC, with CoWoS packaging as the binding constraint. HBM is three suppliers (SK Hynix, Samsung, Micron) and is tight into 2026. - Any Taiwan disruption is unmodelable and total. Short of that, a CoWoS or HBM shortfall caps units regardless of demand.
Short version: the five that matter most, in order – hyperscaler capex durability, China/export policy, custom silicon (TPU/Trainium/ASICs), circular financing of the demand base, and single-point supply chain exposure to TSMC/HBM. Valuation is a sixth, but it's a symptom of the others. Figures below are as of my latest data; verify against the most recent 10-Q.
6. Execution and margins - Annual cadence (Blackwell → Blackwell Ultra → Rubin 2026 → Rubin Ultra 2027) leaves no slack. Blackwell's initial ramp compressed gross margin to ~71% from mid-70s before recovering to ~73–75%; any Rubin slip repeats that. - Data center power availability is becoming the practical gating factor on deployments, which can delay revenue recognition even with orders in hand.
Short version: the five that matter most, in order – hyperscaler capex durability, China/export policy, custom silicon (TPU/Trainium/ASICs), circular financing of the demand base, and single-point supply chain exposure to TSMC/HBM. Valuation is a sixth, but it's a symptom of the others. Figures below are as of my latest data; verify against the most recent 10-Q.
2. China / export controls - China was ~13% of FY25 revenue (~$17B). The H20 ban in April 2025 drove a $4.5B inventory charge and ~$8B of lost quarterly revenue; guidance now embeds essentially zero China data center. - The tail risk isn't just lost revenue – it's Beijing actively steering buyers to Huawei Ascend and probing NVDA on antitrust (Mellanox). A US–China deal that reopens the market is upside; a Blackwell-class chip getting blocked or China banning imports outright is the downside. This is a political variable, not a modeling one.
The biggest live risks are no longer about demand — they're about how the buildout is financed, whether margins hold through the memory-cost squeeze, and a balance sheet that has started carrying the ecosystem's risk. Demand is settled: reported Q2 FY2027 (quarter ended July 26, 2026) revenue was $96.2B, up 106% year over year [Q2 FY2027], and Q3 is guided at [guidance]">$108.0B ±[guidance]">2% [guidance]. Here is what actually threatens the thesis, ranked.
1. Circular / vendor-financed demand (highest priority). NVIDIA is now funding part of its own demand. Three disclosures in the 10-Q filed August 26, 2026 stack up: a new model where NVIDIA guarantees to buy unsold AI-cloud capacity from partners, with six-year commitments totaling [Q3 FY2026]">$36B as of July 26, 2026 [SEC filings]; MOUs to mobilize more than $500B of third-party AI-infrastructure capital, where NVIDIA "may provide limited residual-value support" [SEC filings]; and a $2.9B financing-section payment to Groq alongside non-marketable securities rising from $43.4B to $51.2B [F, quarterly/NVDA]. On top sits the reported ~$250B Nvidia-OpenAI financing/compute deal — the "circular financing" concern that drove a ~4% drop on 2026-07-27; the escalation case is confirmed large vendor financing at aggressive terms [watch]. The financing canary, CoreWeave credit-default swaps (cost of insuring its debt), has recovered from ~[watch]">881bps in December 2025 to ~[watch]">452bps; back above [watch]">600bps would signal renewed neocloud stress [watch].
2. Gross margin trough — scheduled by management itself. Reported GAAP gross margin was [state]">75.0% in Q2 FY2027 [Q2 FY2027], up from [Q1 FY2027]">74.9% in Q1 FY2027 (quarter ended April 26, 2026) [Q1 FY2027] and [Q2 FY2026]">72.4% a year ago [Q2 FY2026]. But guided Q3 margin is [guidance]">74.0% ±50bp [guidance], and management guided Q4 FY2027 to a 71–72% trough on memory costs, with FY2028 at 72–73% [F, Aug-26 guidance]. The 73% "pricing-power" trigger is therefore set to fire on guidance, not surprise [watch]. Because NVIDIA is fabless, memory-price inflation shows up as higher purchased component cost — it's a COGS hit, and the question is how much gets passed through to customers.
3. Supply commitments and a working-capital swell. Supply and capacity commitments more than doubled from $119B to $279B as of July 26, 2026 [SEC filings]. Reported inventories rose to [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B a quarter earlier, receivables jumped to [Q2 FY2027]">$63.1B (DSO ~60 days) from [Q1 FY2027]">$40.7B (~45 days), and total debt went to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B after [Q2 FY2027]">$24.9B of issuance [Q2 FY2027] [Q1 FY2027]. Memory suppliers' inventories have dropped below 10 days, signaling tight HBM supply [risk], and Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" [SEC filings]. A new standalone risk factor on indebtedness and covenants appeared in the filing [SEC filings]. If demand estimates prove inaccurate, $279B of commitments becomes an inventory write-down problem.
4. Hyperscaler capex and customer concentration. A coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL is "THE highest-signal bear trigger" [watch]. One direct customer was [guidance]">16% of Q2 revenue; three were [guidance]">16%, [SEC filings]">15% and [SEC filings]">13% for the first half [SEC filings]. The 10-Q also flags that customers may postpone purchases due to lack of data-center infrastructure or capital, and that AI clouds lack long-term contracts, with land and shell now added to the bottleneck list [SEC filings].
5. China and export controls. The PRC blocked H20 sales despite US licenses, so licensed China revenue never materialized [SEC filings]; Q3 guidance explicitly excludes any China Data Center compute revenue [guidance]. New risk language says foreign governments may steer customers to competitors and that export rules trigger foreign inquiries [SEC filings]. Huawei has released an advanced high-end chip [risk]. A policy thaw is upside optionality; tightening removes even that [watch].
6. Foundry and supplier fragility. Dependence on third-party foundries limits control over yields and delivery, and supplier stress is live — Wistron raised $1.47B for its AI-server ramp, shares falling on dilution [risk].
7. Valuation and relative-value stretch. The comps trigger has fired: peer-median P/S implies ~[state]">-69.5% downside, past the [watch]">-65% threshold [watch] [state]. The reverse DCF (modeled) implies [state]">29.9% annual FCF growth for a decade, under the [watch]">32% ceiling [watch]. Forward P/E is [state]">24.9x (consensus) [state] — cheap on its own history, but the premium to peers is widening.
8. Secondary items. Antitrust probes worldwide into GPU/AI competition [risk]; Hugging Face acquisition integration [risk]; insiders net-sold 4,691,358 shares over six months with zero open-market buys [state]; a board member sold a $410M stake [risk]; Piotroski F-score of 4 with a [state]">10.9% accrual ratio [state] — earnings quality looks weaker than headline growth suggests.
What to watch: Q3 FY2027 gross margin vs the [guidance]">74.0% guide and any Q4 revision below [state]">71%; OpenAI deal confirmation and its structure (equity vs vendor financing); CoreWeave CDS vs [watch]">600bps; hyperscaler 2027 capex guides at October–November earnings.
Bottom line: Demand risk has been replaced by financing risk — NVIDIA now holds [Q3 FY2026]">$36B in capacity guarantees, $279B in supply commitments and [Q3 FY2026]">$33.4B in debt while guiding its own margin down ~350bp over two quarters; the thesis survives if the buildout stays externally funded and powerable, and breaks if the capital providers blink.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
5. China and export controls. The PRC blocked H20 sales despite US licenses, so licensed China revenue never materialized ; Q3 guidance explicitly excludes any China Data Center compute revenue . New risk language says foreign governments may steer customers to competitors and that export rules trigger foreign inquiries . Huawei has released an advanced high-end chip . A policy thaw is upside optionality; tightening removes even that .
4. Hyperscaler capex and customer concentration. A coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL is "THE highest-signal bear trigger" . One direct customer was 16% of Q2 revenue; three were 16%, 15% and 13% for the first half . The 10-Q also flags that customers may postpone purchases due to lack of data-center infrastructure or capital, and that AI clouds lack long-term contracts, with land and shell now added to the bottleneck list .
3. Supply commitments and a working-capital swell. Supply and capacity commitments more than doubled from $119B to $279B as of July 26, 2026 . Reported inventories rose to $31.6B from $25.8B a quarter earlier, receivables jumped to $63.1B (DSO ~60 days) from $40.7B (~45 days), and total debt went to $33.4B from $8.5B after $24.9B of issuance Q2 FY2027 Q1 FY2027. Memory suppliers' inventories have dropped below 10 days, signaling tight HBM supply , and Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" . A new standalone risk factor on indebtedness and covenants appeared in the filing . If demand estimates prove inaccurate, $279B of commitments becomes an inventory write-down problem.
1. Circular / vendor-financed demand (highest priority). NVIDIA is now funding part of its own demand. Three disclosures in the 10-Q filed August 26, 2026 stack up: a new model where NVIDIA guarantees to buy unsold AI-cloud capacity from partners, with six-year commitments totaling $36B as of July 26, 2026 ; MOUs to mobilize more than $500B of third-party AI-infrastructure capital, where NVIDIA "may provide limited residual-value support" ; and a $2.9B financing-section payment to Groq alongside non-marketable securities rising from $43.4B to $51.2B . On top sits the reported ~$250B Nvidia-OpenAI financing/compute deal – the "circular financing" concern that drove a ~4% drop on 2026-07-27; the escalation case is confirmed large vendor financing at aggressive terms . The financing canary, CoreWeave credit-default swaps (cost of insuring its debt), has recovered from ~881bps in December 2025 to ~452bps; back above 600bps would signal renewed neocloud stress .
The biggest live risks are no longer about demand – they're about how the buildout is financed, whether margins hold through the memory-cost squeeze, and a balance sheet that has started carrying the ecosystem's risk. Demand is settled: reported Q2 FY2027 (quarter ended July 26, 2026) revenue was $96.2B, up 106% year over year Q2 FY2027, and Q3 is guided at $108.0B ±2% . Here is what actually threatens the thesis, ranked.
5. China and export controls. The PRC blocked H20 sales despite US licenses, so licensed China revenue never materialized ; Q3 guidance explicitly excludes any China Data Center compute revenue . New risk language says foreign governments may steer customers to competitors and that export rules trigger foreign inquiries . Huawei has released an advanced high-end chip . A policy thaw is upside optionality; tightening removes even that .
4. Hyperscaler capex and customer concentration. A coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL is "THE highest-signal bear trigger" . One direct customer was 16% of Q2 revenue; three were 16%, 15% and 13% for the first half . The 10-Q also flags that customers may postpone purchases due to lack of data-center infrastructure or capital, and that AI clouds lack long-term contracts, with land and shell now added to the bottleneck list .
8. Secondary items. Antitrust probes worldwide into GPU/AI competition ; Hugging Face acquisition integration ; insiders net-sold 4,691,358 shares over six months with zero open-market buys ; a board member sold a $410M stake ; Piotroski F-score of 4 with a 10.9% accrual ratio – earnings quality looks weaker than headline growth suggests.
3. Supply commitments and a working-capital swell. Supply and capacity commitments more than doubled from $119B to $279B as of July 26, 2026 . Reported inventories rose to $31.6B from $25.8B a quarter earlier, receivables jumped to $63.1B (DSO ~60 days) from $40.7B (~45 days), and total debt went to $33.4B from $8.5B after $24.9B of issuance Q2 FY2027 Q1 FY2027. Memory suppliers' inventories have dropped below 10 days, signaling tight HBM supply , and Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" . A new standalone risk factor on indebtedness and covenants appeared in the filing . If demand estimates prove inaccurate, $279B of commitments becomes an inventory write-down problem.
Short version: Over the last ~7 quarters NVDA has beaten its own revenue guide every time (by $1.5–3B), hit its gross-margin roadmap, and shipped Blackwell/GB300 on the revised schedule. The misses are almost all China-related and externally driven, plus a couple of soft "aspirational" numbers (automotive) that are tracking well below what Jensen floated. Note: my data runs through the Q3 FY26 call (Nov 19, 2025) — check the Q4 FY26 print against the $65B guide.
Delivered
Revised
Missed / soft
Open promises to hold them to
Bottom line: On anything management can control — guide, ramp, margins, cadence — the track record is near-perfect and consistently conservative by $1.5–3B a quarter; the misses are China (policy-driven) and one over-optimistic automotive number, so the real credibility test ahead is whether the $500B visibility claim converts to reported revenue in FY27.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
- Quarterly revenue guidance – beat every quarter: - Q1 FY25: guided $24.0B → $26.0B - Q2 FY25: $28.0B → $30.0B - Q3 FY25: $32.5B → $35.1B - Q4 FY25: $37.5B → $39.3B - Q1 FY26: $43.0B → $44.1B (despite a $4.5B H20 write-down and ~$2.5B of unshipped H20) - Q2 FY26: $45.0B → $46.7B - Q3 FY26: $54.0B → $57.0B (Data Center $51.2B) - Blackwell revenue: In Aug 2024 they promised "several billion dollars" of Blackwell in Q4 FY25 after the mask fix; delivered $11B – the fastest product ramp in company history. By Q3 FY26 GB300 was ~two-thirds of Blackwell revenue, confirming the annual cadence (Blackwell → Blackwell Ultra → Rubin) is holding. - Gross margin path: Told investors margins would trough in the low-70s during the Blackwell ramp and return to "mid-70s" exiting FY26. Non-GAAP GM: 71.3% (Q1 FY26, ex-charge) → 72.7% → 73.6% → Q4 guide 75%. On script. - Networking: Spectrum-X promised as a "multi-billion dollar" line; networking hit $8.2B in Q3 FY26 (+162% y/y), well ahead of that framing. - Sovereign AI: Initially "high single-digit billions" for FY25, raised to "low double-digit billions" – met, and FY26 is tracking to more than double. - Capital return: $50B (Aug 2024) + $60B (Aug 2025) authorizations; ~$37B returned in the first three quarters of FY26.
Short version: Over the last ~7 quarters NVDA has beaten its own revenue guide every time (by $1.5–3B), hit its gross-margin roadmap, and shipped Blackwell/GB300 on the revised schedule. The misses are almost all China-related and externally driven, plus a couple of soft "aspirational" numbers (automotive) that are tracking well below what Jensen floated. Note: my data runs through the Q3 FY26 call (Nov 19, 2025) – check the Q4 FY26 print against the $65B guide.
- Quarterly revenue guidance – beat every quarter: - Q1 FY25: guided $24.0B → $26.0B - Q2 FY25: $28.0B → $30.0B - Q3 FY25: $32.5B → $35.1B - Q4 FY25: $37.5B → $39.3B - Q1 FY26: $43.0B → $44.1B (despite a $4.5B H20 write-down and ~$2.5B of unshipped H20) - Q2 FY26: $45.0B → $46.7B - Q3 FY26: $54.0B → $57.0B (Data Center $51.2B) - Blackwell revenue: In Aug 2024 they promised "several billion dollars" of Blackwell in Q4 FY25 after the mask fix; delivered $11B – the fastest product ramp in company history. By Q3 FY26 GB300 was ~two-thirds of Blackwell revenue, confirming the annual cadence (Blackwell → Blackwell Ultra → Rubin) is holding. - Gross margin path: Told investors margins would trough in the low-70s during the Blackwell ramp and return to "mid-70s" exiting FY26. Non-GAAP GM: 71.3% (Q1 FY26, ex-charge) → 72.7% → 73.6% → Q4 guide 75%. On script. - Networking: Spectrum-X promised as a "multi-billion dollar" line; networking hit $8.2B in Q3 FY26 (+162% y/y), well ahead of that framing. - Sovereign AI: Initially "high single-digit billions" for FY25, raised to "low double-digit billions" – met, and FY26 is tracking to more than double. - Capital return: $50B (Aug 2024) + $60B (Aug 2025) authorizations; ~$37B returned in the first three quarters of FY26.
- Quarterly revenue guidance – beat every quarter: - Q1 FY25: guided $24.0B → $26.0B - Q2 FY25: $28.0B → $30.0B - Q3 FY25: $32.5B → $35.1B - Q4 FY25: $37.5B → $39.3B - Q1 FY26: $43.0B → $44.1B (despite a $4.5B H20 write-down and ~$2.5B of unshipped H20) - Q2 FY26: $45.0B → $46.7B - Q3 FY26: $54.0B → $57.0B (Data Center $51.2B) - Blackwell revenue: In Aug 2024 they promised "several billion dollars" of Blackwell in Q4 FY25 after the mask fix; delivered $11B – the fastest product ramp in company history. By Q3 FY26 GB300 was ~two-thirds of Blackwell revenue, confirming the annual cadence (Blackwell → Blackwell Ultra → Rubin) is holding. - Gross margin path: Told investors margins would trough in the low-70s during the Blackwell ramp and return to "mid-70s" exiting FY26. Non-GAAP GM: 71.3% (Q1 FY26, ex-charge) → 72.7% → 73.6% → Q4 guide 75%. On script. - Networking: Spectrum-X promised as a "multi-billion dollar" line; networking hit $8.2B in Q3 FY26 (+162% y/y), well ahead of that framing. - Sovereign AI: Initially "high single-digit billions" for FY25, raised to "low double-digit billions" – met, and FY26 is tracking to more than double. - Capital return: $50B (Aug 2024) + $60B (Aug 2025) authorizations; ~$37B returned in the first three quarters of FY26.
- Quarterly revenue guidance – beat every quarter: - Q1 FY25: guided $24.0B → $26.0B - Q2 FY25: $28.0B → $30.0B - Q3 FY25: $32.5B → $35.1B - Q4 FY25: $37.5B → $39.3B - Q1 FY26: $43.0B → $44.1B (despite a $4.5B H20 write-down and ~$2.5B of unshipped H20) - Q2 FY26: $45.0B → $46.7B - Q3 FY26: $54.0B → $57.0B (Data Center $51.2B) - Blackwell revenue: In Aug 2024 they promised "several billion dollars" of Blackwell in Q4 FY25 after the mask fix; delivered $11B – the fastest product ramp in company history. By Q3 FY26 GB300 was ~two-thirds of Blackwell revenue, confirming the annual cadence (Blackwell → Blackwell Ultra → Rubin) is holding. - Gross margin path: Told investors margins would trough in the low-70s during the Blackwell ramp and return to "mid-70s" exiting FY26. Non-GAAP GM: 71.3% (Q1 FY26, ex-charge) → 72.7% → 73.6% → Q4 guide 75%. On script. - Networking: Spectrum-X promised as a "multi-billion dollar" line; networking hit $8.2B in Q3 FY26 (+162% y/y), well ahead of that framing. - Sovereign AI: Initially "high single-digit billions" for FY25, raised to "low double-digit billions" – met, and FY26 is tracking to more than double. - Capital return: $50B (Aug 2024) + $60B (Aug 2025) authorizations; ~$37B returned in the first three quarters of FY26.
Scorecard: 31 promises across the last six earnings calls (May 2025 to August 2026), 17 resolved and 14 still pending — and the pattern is unusually clean. Every product, revenue and capital-return promise in the record was delivered or beaten; every operating-expense promise was revised upward, four times running. No outright misses appear in the resolved set [transcripts]. (Of the 17 resolved items, eight are itemized in front of me; the rest aren't broken out.)
Delivered (4 itemized)
Revised / withdrawn (4 itemized — all one theme)
Operating-expense growth guidance has been raised every time it's been restated:
For scale, guided Q3 FY2027 GAAP opex is ~[guidance]">$9.2B against guided revenue of [guidance]">$108.0B ±[guidance]">2% [guidance] — roughly 8.5% of sales, so the creep hasn't dented profitability (reported FY2026 net margin [state]">55.6%) [state]. It does tell you management systematically under-forecasts its own spending, which is worth remembering when reading the next cost line they guide.
Missed: none — but the next margin promise is already the live one
The mid-70s margin promise was kept; the successor promise is a managed decline. Reported Q2 FY2027 GAAP gross margin was [state]">75.0%, up from [Q1 FY2027]">74.9% in Q1 FY2027 and [Q2 FY2026]">72.4% a year ago [Q2 FY2027]. Guidance (not yet reported) is [guidance]">74.0% ±50bp for Q3 FY2027 [guidance], a trough of 71–72% in Q4 FY2027, and 72–73% for FY2028, because memory price increases "exceeded prior expectations and are headed even higher" [earnings recap agent]. The gross-margin watch trigger sits at 73% and is scheduled to fire on management's own Q4 guide [watch]. The question is no longer "will they hit mid-70s" but "will the trough really stop at [state]">71%," given the opex track record of guiding low.
One other near-miss against the Street rather than their own word: the Q3 revenue guide of [guidance]">$108.0B landed 0.6% below the $108.65B consensus, so the Aug 26 print scored as a beat (EPS $2.22 vs $2.09 consensus; revenue $96.22B vs $92.27B) but not a beat-and-raise — the streak of double beats is 7-for-7 [earnings recap agent]. Shares still rose [earnings recap agent]">8.7% on [earnings recap agent]">2.6x normal volume [earnings recap agent].
Pending promises to hold them to
What to watch: (1) Nov 18 print — Q3 revenue vs [guidance]">$108.0B guide and gross margin vs [guidance]">74% ±50bp; consensus EPS is [estimate]">$2.47 [estimate]. (2) Whether the Q4 FY2027 margin trough guide holds at 71–72% or is re-lowered the way opex was. (3) Full-year opex growth vs "low 50s" — a fifth upward revision would confirm the pattern.
Bottom line: Management keeps its product, revenue and shareholder promises and consistently breaks its cost promises upward; that makes the guided 71–72% margin trough the promise most worth doubting, and the [earnings recap agent]">70% FY2028 growth guide the one most likely to be conservative.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Operating-expense growth guidance has been raised every time it's been restated: - FY2026: "mid-thirties" (Feb 2025, reaffirmed May 2025) → "high thirties" (Aug 2025) . - FY2027: "low 40s" (Feb 2026) → "upper forties" (May 2026) → "low 50s" (Aug 2026), attributed to "a broadening of our product portfolio and further increase in the usage of AI tools" .
Operating-expense growth guidance has been raised every time it's been restated: - FY2026: "mid-thirties" (Feb 2025, reaffirmed May 2025) → "high thirties" (Aug 2025) . - FY2027: "low 40s" (Feb 2026) → "upper forties" (May 2026) → "low 50s" (Aug 2026), attributed to "a broadening of our product portfolio and further increase in the usage of AI tools" .
- Vera Rubin production shipments in 2H calendar 2026. Promised Feb 25, 2026; confirmed Aug 26, 2026: "We commenced production shipments of Vera Rubin earlier this month, having already received purchase orders from every major hyperscaler, AI cloud and system OEM." The 10-Q corroborates shipments began, while noting supply constraints persist . - Capital return. Promised Feb 2026; Q2 FY2027 (quarter ended July 26, 2026) delivered a record $26B returned – $20B buybacks, $6B dividends at $0.25/share . Reported fiscal-year-to-date buybacks are $39.0B and dividends $6.3B – but note that $24.9B of debt was issued in the same period, taking total debt to $33.4B from $8.5B a quarter earlier, so the return was partly debt-funded Q2 FY2027. - Exit FY2026 with mid-70s non-GAAP gross margin. Promised Aug 2025; Q4 FY2026 (quarter ended January 2026) printed 75% GAAP / 75.2% non-GAAP as Blackwell ramped . - >$20B Sovereign AI revenue in FY2026. Beaten by half: the business "more than tripled year-over-year" to over $30B .
- Vera Rubin production shipments in 2H calendar 2026. Promised Feb 25, 2026; confirmed Aug 26, 2026: "We commenced production shipments of Vera Rubin earlier this month, having already received purchase orders from every major hyperscaler, AI cloud and system OEM." The 10-Q corroborates shipments began, while noting supply constraints persist . - Capital return. Promised Feb 2026; Q2 FY2027 (quarter ended July 26, 2026) delivered a record $26B returned – $20B buybacks, $6B dividends at $0.25/share . Reported fiscal-year-to-date buybacks are $39.0B and dividends $6.3B – but note that $24.9B of debt was issued in the same period, taking total debt to $33.4B from $8.5B a quarter earlier, so the return was partly debt-funded Q2 FY2027. - Exit FY2026 with mid-70s non-GAAP gross margin. Promised Aug 2025; Q4 FY2026 (quarter ended January 2026) printed 75% GAAP / 75.2% non-GAAP as Blackwell ramped . - >$20B Sovereign AI revenue in FY2026. Beaten by half: the business "more than tripled year-over-year" to over $30B .
- Vera Rubin production shipments in 2H calendar 2026. Promised Feb 25, 2026; confirmed Aug 26, 2026: "We commenced production shipments of Vera Rubin earlier this month, having already received purchase orders from every major hyperscaler, AI cloud and system OEM." The 10-Q corroborates shipments began, while noting supply constraints persist . - Capital return. Promised Feb 2026; Q2 FY2027 (quarter ended July 26, 2026) delivered a record $26B returned – $20B buybacks, $6B dividends at $0.25/share . Reported fiscal-year-to-date buybacks are $39.0B and dividends $6.3B – but note that $24.9B of debt was issued in the same period, taking total debt to $33.4B from $8.5B a quarter earlier, so the return was partly debt-funded Q2 FY2027. - Exit FY2026 with mid-70s non-GAAP gross margin. Promised Aug 2025; Q4 FY2026 (quarter ended January 2026) printed 75% GAAP / 75.2% non-GAAP as Blackwell ramped . - >$20B Sovereign AI revenue in FY2026. Beaten by half: the business "more than tripled year-over-year" to over $30B .
- FY2028 revenue growth ~70%, explicitly supply-constrained; management estimates true demand growth at 100% and says the 70% reflects confirmed capacity . Consensus FY2028 revenue is ~$686B, revised up 1% in the last week . - Q3 FY2027: $108.0B ±2% revenue, 74% ±50bp gross margin, ~$9.0B non-GAAP opex, 16–18% tax rate – China Data Center compute is excluded as a conservative assumption . - Vera Rubin as "the fastest ramp in history" – with reported inventories at $31.6B (from $25.8B in Q1 FY2027) and receivables at $63.1B, DSO ~60 days vs ~45 a quarter earlier Q2 FY2027, the working-capital lines will show whether the ramp is shipping and getting paid. - Nearly $50B of exposure to pre-profit Frontier AI labs, which management argues is manageable because compute is redeployable ; the reported ~$250B OpenAI financing/compute arrangement is unconfirmed and on watch .
Scorecard: 31 promises across the last six earnings calls (May 2025 to August 2026), 17 resolved and 14 still pending – and the pattern is unusually clean. Every product, revenue and capital-return promise in the record was delivered or beaten; every operating-expense promise was revised upward, four times running. No outright misses appear in the resolved set . (Of the 17 resolved items, eight are itemized in front of me; the rest aren't broken out.)
Comparison: Q3 FY26 10-Q (quarter ended Oct 26, 2025, filed Nov 19, 2025) vs. Q2 FY26 10-Q (quarter ended Jul 27, 2025). Verify against the filing if a newer one has landed since.
Headline changes
New/changed disclosures that matter for the thesis
Balance-sheet build ahead of demand. Inventory rose to ~[Q3 FY2026]">$19.8B from ~[Q2 FY2026]">$15.0B, and supply purchase commitments/prepaids climbed to ~$50B from ~$44B. This is the company putting its own capital behind the $500B Blackwell+Rubin visibility claim. Bull read: locked-in supply. Bear read: the inventory write-down risk (see the $4.5B H20 charge in Q1) scales with the commitment.
Receivables jumped faster than revenue. AR ~[Q3 FY2026]">$33B vs. ~$23B, pushing DSO up meaningfully. Management attributed it to shipment timing late in the quarter. It's the single line I'd watch for normalization in Q4 — if it doesn't come back down, the "demand is sold before it's built" narrative weakens.
Customer concentration ticked up and shifted. Two direct customers ≈22% and ≈[SEC filings]">15% of revenue (vs. ~23%/~[guidance]">16% in Q2), with indirect hyperscaler exposure acknowledged as higher. New language flags that customer capex plans and their own custom-silicon efforts are a risk — an explicit nod to TPU/Trainium.
China effectively zeroed out. Q3 H20 revenue was immaterial (~$50M); Q4 guidance assumes no China data center compute. The 10-Q dropped the "if licenses arrive" upside framing and added disclosure on Chinese government discouragement of purchases. China is now optionality, not base case — that's actually thesis-positive because the guide doesn't need it.
Equity investments in customers/partners. Non-marketable investments expanded materially (OpenAI LOI up to $100B, Anthropic $10B, Intel $5B, Nokia [Q3 FY2026]">$1B announced in-quarter or just after). Risk-factor language was broadened to cover investing in entities that are also customers. This is the "circular financing" issue — the 10-Q now explicitly acknowledges revenue may depend partly on capital NVDA itself supplies. Not disqualifying, but it's the disclosure most likely to be cited if the AI capex cycle wobbles.
Legal/regulatory. Securities class action (revived after the Dec 2024 Supreme Court dismissal) remains ongoing; new references to potential tariff/export-control changes and China antitrust review of the Mellanox deal.
What didn't change Cash and securities ~$60B, negligible net debt, opex growing ~mid-teens q/q, no change to segment reporting.
Bottom line: The Q3 10-Q shows a business accelerating with rising margins and a fully de-risked China guide, but the new disclosures to underwrite are the working-capital build (AR + inventory + commitments up ~[Q3 FY2026]">$20B combined) and the formalized risk language around investing in your own customers — that's where the bear case now lives, not in demand.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
3. Customer concentration ticked up and shifted. Two direct customers ≈22% and ≈15% of revenue (vs. ~23%/~16% in Q2), with indirect hyperscaler exposure acknowledged as higher. New language flags that customer capex plans and their own custom-silicon efforts are a risk – an explicit nod to TPU/Trainium.
6. Legal/regulatory. Securities class action (revived after the Dec 2024 Supreme Court dismissal) remains ongoing; new references to potential tariff/export-control changes and China antitrust review of the Mellanox deal.
The short version: the 10-Q filed 2026-08-26 (for Q2 FY2027, quarter ended July 26, 2026) shows a company that has stopped simply selling AI infrastructure and started financing and underwriting it. Of the 11 material disclosure changes versus the May 20 filing, three are new and matter most for the thesis: a $279 billion supply commitment (up from $119 billion), a [Q3 FY2026]">$36 billion guarantee to buy unsold cloud capacity from AI cloud partners, and MOUs to mobilize $500 billion of third-party capital with optional residual-value support from NVIDIA. Together they extend the balance sheet into the demand side of its own market — which is exactly the "circular financing" worry that hit the stock in July. [SEC filings]
1. Supply commitments more than doubled. Reported supply and capacity commitments went from $119 billion at the prior quarter to $279 billion as of July 26, 2026 — a $160 billion increase in one quarter — to meet future demand, with the filing explicitly warning that customers may postpone purchases due to a lack of data center infrastructure, capital constraints, or slower adoption. [SEC filings] This is the single biggest thesis-relevant number in the filing: NVIDIA is fabless, so this is not capex on its own plants but purchase obligations to foundries and component suppliers. If demand rolls over, the risk shows up as inventory and purchase-commitment charges against gross margin. Inventories were already a reported [Q2 FY2027]">$31.6 billion at Q2 FY2027, up from [Q1 FY2027]">$25.8 billion in Q1 FY2027 (quarter ended April 26, 2026) and [Q2 FY2026]">$15.0 billion in Q2 FY2026 (quarter ended July 27, 2025) [Q2 FY2027] [Q1 FY2027] [Q2 FY2026].
2. A new business model with contingent obligations. NVIDIA introduced a model with select AI cloud partners under which it commits to purchase unsold capacity; those commitments, typically six years long, totaled a reported [Q3 FY2026]">$36 billion as of July 26, 2026, decreasing as third parties or NVIDIA's own R&D uses the capacity. [SEC filings] A separate new risk factor confirms NVIDIA guarantees these purchases if capacity goes unsold — a material downside if third-party demand underperforms. [SEC filings] Read plainly: NVIDIA is now a backstop customer for the "neoclouds" that buy its chips. That links its P&L to their health — the reason the CoreWeave CDS canary (credit-default-swap spread — the cost of insuring its debt) is on the watch list, having recovered from ~[watch]">881bps in Dec-2025 to ~[watch]">452bps, with [watch]">600bps the renewed-stress line. [watch]
3. $500 billion of third-party capital. In August 2026 NVIDIA signed MOUs with large capital providers to set up independent financing platforms mobilizing more than $500 billion for AI infrastructure; funding is to be underwritten by those providers, but "at our option" NVIDIA may provide limited residual-value support for specific projects, and the arrangements may not become definitive. [SEC filings] Positive for addressable market, but it is the third leg of the same theme: NVIDIA capital, directly or via guarantees, standing behind demand for NVIDIA products. The reported ~$250B OpenAI compute/financing deal that drove a ~4% drop on 2026-07-27 remains unconfirmed and is on watch for structure — equity stake vs vendor financing. [watch]
4. Vera Rubin shipping, but supply constrained. The next-generation architecture began production shipments in Q3 FY2027; NVIDIA will ship Blackwell and Rubin concurrently and is "currently experiencing certain supply constraints," with demand estimates that "can be inaccurate and create volatility." [SEC filings] Data Center revenue was a reported $89.0 billion, up 117% y/y and [guidance]">18% sequentially on Blackwell Ultra; hyperscale more than doubled y/y and ACIE grew [SEC filings]">138% y/y. [SEC filings] The margin frame around this ramp: guided Q3 FY2027 GAAP gross margin of [guidance]">74.0% ±50bp on guided revenue of [guidance]">$108.0 billion ±[guidance]">2% [guidance], but management also guided Q4 FY2027 gross margin down to 71–72% on memory cost inflation — which means the 73% gross-margin watch trigger is scheduled to fire on management's own outlook, against a reported [state]">75.0% in Q2 FY2027. [watch] [Q2 FY2027]
5. China: optionality shrank further. The filing confirms the PRC government blocked H20 sales despite U.S. licenses, so NVIDIA could not sell that inventory — reversing the earlier expectation of a licensed China stream. The specific China antitrust preliminary finding (tied to Mellanox commitments) was replaced by generalized export-control language, and two new risks were added: export rules triggering foreign-government inquiries, and foreign governments steering customers to competitors. [SEC filings] Q3 guidance already excludes any Data Center compute revenue from China, so the base case doesn't depend on it [guidance] — but the filing removes even the optionality the watch list flags. [watch]
6. Concentration, receivables and debt. Customer concentration eased on paper: one direct customer at [guidance]">16% of Q2 revenue versus three at 21%/17%/[guidance]">16% in Q1; first-half concentration was [guidance]">16%/[SEC filings]">15%/[SEC filings]">13%. [SEC filings] But reported receivables jumped to [Q2 FY2027]">$63.1 billion (DSO ~60 days) from [Q1 FY2027]">$40.7 billion (~45 days) in Q1 — a 55% increase against [guidance]">18% revenue growth ($96.2B vs $81.6B) — the highest DSO in the five-quarter tape. [Q2 FY2027] [Q1 FY2027] Total debt rose to a reported [Q3 FY2026]">$33.4 billion from [Q1 FY2026]">$8.5 billion, on [Q2 FY2027]">$24.9 billion issued fiscal-year-to-date, funding [Q2 FY2027]">$39.0 billion of buybacks and [Q2 FY2027]">$6.3 billion of dividends [Q2 FY2027]; a standalone indebtedness risk factor was added, and the marketable-securities description narrowed to U.S. government-only holdings. [SEC filings] Credit capacity is still rated strong with 0 warnings, so this is a posture change, not a stress signal. [company analyst; changes only where store content moved]
Why it matters for the thesis. The debate is currently bear-leading on the crux that the price already embeds ~29% annual FCF growth for a decade (modeled reverse-DCF implied growth: [state]">29.9%). [debate agent] [state] Every new disclosure raises the cost of a demand stumble — $279 billion of supply obligations plus [Q3 FY2026]">$36 billion of capacity guarantees (about $315 billion combined) now sit against a guided margin step-down and a 60-day DSO — while the stock's tape has done nothing but reward the beat (+8.7% on 2.6x volume). [earnings recap]
What to watch: (1) Q4 FY2027 gross margin — inside 71–72% or below, and whether memory inflation persists past one quarter; (2) the [Q3 FY2026]">$36 billion capacity commitment and $279 billion supply figure in the next 10-Q — growing or drawing down; (3) whether the $500 billion MOUs become definitive and how much residual-value support NVIDIA actually takes on; (4) 2027 hyperscaler capex guidance across MSFT/AMZN/META/GOOGL. [watch]
Bottom line: The 10-Q confirms demand is real and accelerating, but NVIDIA is now financing, guaranteeing and pre-buying for that demand on a scale ($279B supply, [Q3 FY2026]">$36B capacity, $500B MOUs) that converts a cyclical slowdown from a revenue miss into a balance-sheet event — at a price that already assumes none occurs.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
3. $500 billion of third-party capital. In August 2026 NVIDIA signed MOUs with large capital providers to set up independent financing platforms mobilizing more than $500 billion for AI infrastructure; funding is to be underwritten by those providers, but "at our option" NVIDIA may provide limited residual-value support for specific projects, and the arrangements may not become definitive. Positive for addressable market, but it is the third leg of the same theme: NVIDIA capital, directly or via guarantees, standing behind demand for NVIDIA products. The reported ~$250B OpenAI compute/financing deal that drove a ~4% drop on 2026-07-27 remains unconfirmed and is on watch for structure – equity stake vs vendor financing.
The short version: the 10-Q filed 2026-08-26 (for Q2 FY2027, quarter ended July 26, 2026) shows a company that has stopped simply selling AI infrastructure and started financing and underwriting it. Of the 11 material disclosure changes versus the May 20 filing, three are new and matter most for the thesis: a $279 billion supply commitment (up from $119 billion), a $36 billion guarantee to buy unsold cloud capacity from AI cloud partners, and MOUs to mobilize $500 billion of third-party capital with optional residual-value support from NVIDIA. Together they extend the balance sheet into the demand side of its own market – which is exactly the "circular financing" worry that hit the stock in July.
The short version: the 10-Q filed 2026-08-26 (for Q2 FY2027, quarter ended July 26, 2026) shows a company that has stopped simply selling AI infrastructure and started financing and underwriting it. Of the 11 material disclosure changes versus the May 20 filing, three are new and matter most for the thesis: a $279 billion supply commitment (up from $119 billion), a $36 billion guarantee to buy unsold cloud capacity from AI cloud partners, and MOUs to mobilize $500 billion of third-party capital with optional residual-value support from NVIDIA. Together they extend the balance sheet into the demand side of its own market – which is exactly the "circular financing" worry that hit the stock in July.
6. Concentration, receivables and debt. Customer concentration eased on paper: one direct customer at 16% of Q2 revenue versus three at 21%/17%/16% in Q1; first-half concentration was 16%/15%/13%. But reported receivables jumped to $63.1 billion (DSO ~60 days) from $40.7 billion (~45 days) in Q1 – a 55% increase against 18% revenue growth ($96.2B vs $81.6B) – the highest DSO in the five-quarter tape. Q2 FY2027 Q1 FY2027 Total debt rose to a reported $33.4 billion from $8.5 billion, on $24.9 billion issued fiscal-year-to-date, funding $39.0 billion of buybacks and $6.3 billion of dividends Q2 FY2027; a standalone indebtedness risk factor was added, and the marketable-securities description narrowed to U.S. government-only holdings. Credit capacity is still rated strong with 0 warnings, so this is a posture change, not a stress signal.
5. China: optionality shrank further. The filing confirms the PRC government blocked H20 sales despite U.S. licenses, so NVIDIA could not sell that inventory – reversing the earlier expectation of a licensed China stream. The specific China antitrust preliminary finding (tied to Mellanox commitments) was replaced by generalized export-control language, and two new risks were added: export rules triggering foreign-government inquiries, and foreign governments steering customers to competitors. Q3 guidance already excludes any Data Center compute revenue from China, so the base case doesn't depend on it – but the filing removes even the optionality the watch list flags.
5. China: optionality shrank further. The filing confirms the PRC government blocked H20 sales despite U.S. licenses, so NVIDIA could not sell that inventory – reversing the earlier expectation of a licensed China stream. The specific China antitrust preliminary finding (tied to Mellanox commitments) was replaced by generalized export-control language, and two new risks were added: export rules triggering foreign-government inquiries, and foreign governments steering customers to competitors. Q3 guidance already excludes any Data Center compute revenue from China, so the base case doesn't depend on it – but the filing removes even the optionality the watch list flags.
4. Vera Rubin shipping, but supply constrained. The next-generation architecture began production shipments in Q3 FY2027; NVIDIA will ship Blackwell and Rubin concurrently and is "currently experiencing certain supply constraints," with demand estimates that "can be inaccurate and create volatility." Data Center revenue was a reported $89.0 billion, up 117% y/y and 18% sequentially on Blackwell Ultra; hyperscale more than doubled y/y and ACIE grew 138% y/y. The margin frame around this ramp: guided Q3 FY2027 GAAP gross margin of 74% ±50bp on guided revenue of $108.0 billion ±2% , but management also guided Q4 FY2027 gross margin down to 71–72% on memory cost inflation – which means the 73% gross-margin watch trigger is scheduled to fire on management's own outlook, against a reported 75% in Q2 FY2027. Q2 FY2027
6. Concentration, receivables and debt. Customer concentration eased on paper: one direct customer at 16% of Q2 revenue versus three at 21%/17%/16% in Q1; first-half concentration was 16%/15%/13%. But reported receivables jumped to $63.1 billion (DSO ~60 days) from $40.7 billion (~45 days) in Q1 – a 55% increase against 18% revenue growth ($96.2B vs $81.6B) – the highest DSO in the five-quarter tape. Q2 FY2027 Q1 FY2027 Total debt rose to a reported $33.4 billion from $8.5 billion, on $24.9 billion issued fiscal-year-to-date, funding $39.0 billion of buybacks and $6.3 billion of dividends Q2 FY2027; a standalone indebtedness risk factor was added, and the marketable-securities description narrowed to U.S. government-only holdings. Credit capacity is still rated strong with 0 warnings, so this is a posture change, not a stress signal.
5. China: optionality shrank further. The filing confirms the PRC government blocked H20 sales despite U.S. licenses, so NVIDIA could not sell that inventory – reversing the earlier expectation of a licensed China stream. The specific China antitrust preliminary finding (tied to Mellanox commitments) was replaced by generalized export-control language, and two new risks were added: export rules triggering foreign-government inquiries, and foreign governments steering customers to competitors. Q3 guidance already excludes any Data Center compute revenue from China, so the base case doesn't depend on it – but the filing removes even the optionality the watch list flags.
Headline number: two direct customers were ~37% of revenue in the most recent quarter (Customer A ~22%, Customer B ~15%), and effectively 100% of leading-edge wafers and CoWoS packaging come from TSMC, mostly in Taiwan. That's the shape of the risk: a fabless company with one foundry, three memory vendors, and a customer base of roughly six hyperscalers/labs sitting behind a handful of ODMs.
Upstream: where the single points of failure are
Midstream: ODM/system builders
Direct "Customers A and B" are almost certainly system integrators (Foxconn, Quanta, Dell, Supermicro), not end users. Foxconn assembles the majority of GB200/GB300 NVL72 racks (Taiwan, Mexico, new Houston site). Supermicro was a 2024 scare (auditor resignation, delisting risk) and has stabilized. Direct customer concentration has risen from ~13% for the top customer in FY24 to ~22% now—reflecting the shift from boards to full racks, which funnels volume through fewer integrators.
Downstream: end-customer concentration
Geographic/regulatory overlay
Taiwan hosts the fab, packaging, and much of the rack assembly. Section 232 semiconductor tariffs remain an open threat to landed cost; Arizona fab/packaging and Foxconn/Wistron Texas capacity are the mitigants, but they won't be material before 2027.
Bottom line: NVDA's supply chain is a single-foundry, single-island, three-memory-vendor structure feeding ~six end customers who are now simultaneously its biggest buyers, its co-investors, and its emerging competitors—demand visibility through 2026 is excellent, but TSMC/Taiwan, HBM pricing, and customer silicon are the three risks that actually matter.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
- Foundry (TSMC, ~100%). Blackwell/Blackwell Ultra on 4NP, Rubin on N3P. No second source at the leading edge; Samsung isn't qualified and the $5B Intel stake (Sept 2025) is about x86/NVLink integration, not foundry. Mitigant: first Blackwell wafers came off TSMC Arizona in Oct 2025, but Arizona is a small slice of volume and still ships back to Taiwan for packaging. - Advanced packaging (CoWoS-L, TSMC). This was the binding constraint through 2024-25. TSMC roughly doubled CoWoS to ~70-80k wafers/month by end-2025, targeting ~100k+ in 2026; NVDA is believed to take well over half of it. Amkor Arizona and SPIL/ASE are the diversification story, but not meaningful until 2027+. - HBM (SK Hynix dominant, Micron growing, Samsung late). SK Hynix supplied the majority of HBM3E; Micron is in GB300 and Samsung finally got HBM3E qualified in 2025 and is sampling HBM4 for Rubin. New risk in 2025-26: a broad DRAM shortage–HBM is crowding out conventional DRAM and contract prices are up sharply, which pressures gross margin (guided ~mid-70s) and could gate unit volumes. - Substrates / passives / power: Ibiden and Unimicron (ABF substrates), Vertiv and CoolIT (liquid cooling), Amphenol (NVLink copper), Coherent/Lumentum (optics). Less concentrated per supplier, but rack-scale systems mean any one shortage stalls a $3M+ NVL72 rack. - Balance-sheet tell: purchase commitments and supply obligations are ~$50B, and inventory ~$20B–NVDA is prepaying and pre-committing to lock capacity, which is a hedge against shortages but a write-down risk if demand mix shifts (see the $4.5B H20 charge below).
- Foundry (TSMC, ~100%). Blackwell/Blackwell Ultra on 4NP, Rubin on N3P. No second source at the leading edge; Samsung isn't qualified and the $5B Intel stake (Sept 2025) is about x86/NVLink integration, not foundry. Mitigant: first Blackwell wafers came off TSMC Arizona in Oct 2025, but Arizona is a small slice of volume and still ships back to Taiwan for packaging. - Advanced packaging (CoWoS-L, TSMC). This was the binding constraint through 2024-25. TSMC roughly doubled CoWoS to ~70-80k wafers/month by end-2025, targeting ~100k+ in 2026; NVDA is believed to take well over half of it. Amkor Arizona and SPIL/ASE are the diversification story, but not meaningful until 2027+. - HBM (SK Hynix dominant, Micron growing, Samsung late). SK Hynix supplied the majority of HBM3E; Micron is in GB300 and Samsung finally got HBM3E qualified in 2025 and is sampling HBM4 for Rubin. New risk in 2025-26: a broad DRAM shortage–HBM is crowding out conventional DRAM and contract prices are up sharply, which pressures gross margin (guided ~mid-70s) and could gate unit volumes. - Substrates / passives / power: Ibiden and Unimicron (ABF substrates), Vertiv and CoolIT (liquid cooling), Amphenol (NVLink copper), Coherent/Lumentum (optics). Less concentrated per supplier, but rack-scale systems mean any one shortage stalls a $3M+ NVL72 rack. - Balance-sheet tell: purchase commitments and supply obligations are ~$50B, and inventory ~$20B–NVDA is prepaying and pre-committing to lock capacity, which is a hedge against shortages but a write-down risk if demand mix shifts (see the $4.5B H20 charge below).
- Large cloud service providers are roughly half of data-center revenue; two indirect customers (hyperscalers) each exceed 10%. Combined 2025 capex for MSFT/AMZN/GOOGL/META was ~$400B, with 2026 guides pointing well above that. - What's changed: (1) NVDA is now financing its own demand–up to $100B into OpenAI (10 GW), $10B into Anthropic, ~7% of CoreWeave plus a $6.3B capacity backstop. This deepens ties but raises legitimate circularity questions. (2) Customers are diversifying away: OpenAI signed 6 GW with AMD (MI450) and a 10 GW custom-ASIC deal with Broadcom; Anthropic committed to up to 1M Google TPUs and Meta reportedly is evaluating TPUs; Amazon Trainium 3 and Microsoft Maia keep ramping. None dents near-term numbers–NVDA claims $500B of Blackwell+Rubin visibility through CY2026–but it caps pricing power at the margin. - China: went from ~13% of revenue to effectively zero in guidance after the April 2025 H20 license requirement ($4.5B write-down, ~$8B lost Q2 revenue). Washington then allowed H20 sales with a 15% revenue share, Beijing told firms not to buy, and in Dec 2025 the administration cleared H200 exports with a 25% cut. Treat any China contribution as option value, not base case.
- Large cloud service providers are roughly half of data-center revenue; two indirect customers (hyperscalers) each exceed 10%. Combined 2025 capex for MSFT/AMZN/GOOGL/META was ~$400B, with 2026 guides pointing well above that. - What's changed: (1) NVDA is now financing its own demand–up to $100B into OpenAI (10 GW), $10B into Anthropic, ~7% of CoreWeave plus a $6.3B capacity backstop. This deepens ties but raises legitimate circularity questions. (2) Customers are diversifying away: OpenAI signed 6 GW with AMD (MI450) and a 10 GW custom-ASIC deal with Broadcom; Anthropic committed to up to 1M Google TPUs and Meta reportedly is evaluating TPUs; Amazon Trainium 3 and Microsoft Maia keep ramping. None dents near-term numbers–NVDA claims $500B of Blackwell+Rubin visibility through CY2026–but it caps pricing power at the margin. - China: went from ~13% of revenue to effectively zero in guidance after the April 2025 H20 license requirement ($4.5B write-down, ~$8B lost Q2 revenue). Washington then allowed H20 sales with a 15% revenue share, Beijing told firms not to buy, and in Dec 2025 the administration cleared H200 exports with a 25% cut. Treat any China contribution as option value, not base case.
Headline number: two direct customers were ~37% of revenue in the most recent quarter (Customer A ~22%, Customer B ~15%), and effectively 100% of leading-edge wafers and CoWoS packaging come from TSMC, mostly in Taiwan. That's the shape of the risk: a fabless company with one foundry, three memory vendors, and a customer base of roughly six hyperscalers/labs sitting behind a handful of ODMs.
The short version: NVIDIA's chain is concentrated at every layer, and the layer under the most stress right now is memory — not foundry, not demand. NVIDIA is fabless: it designs the GPUs and depends on third-party foundries, a handful of memory makers, and Taiwanese ODMs to build the systems. That structure means a supply shock hits NVIDIA's cost of goods (via supplier pricing) and its ability to ship, rather than its own factories. The near-term evidence is exactly that pattern: memory suppliers are running out of inventory, and management has already guided gross margin down because of it.
Upstream — foundry and memory (single-source, tight)
The FY2026 10-K names both the foundry and the memory suppliers in its manufacturing section, and the risk factors flag dependence on third-party foundries as high: NVIDIA lacks control over "quantity, quality, yields, and delivery schedules" [risk]. That is a standing risk, not a new one.
What is new is memory. Samsung Electronics and SK Hynix inventories have dropped below 10 days, described as a critical shortage of HBM/DRAM (the high-bandwidth memory stacked next to every data-center GPU) that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" [supply-chain agent]. A second report corroborates a "hard capacity ceiling near-term" for module output [supply-chain agent]. The risk register now marks supply chain & manufacturing as LIVE elevated on this headline [risk].
This is already showing up in the P&L guidance. The latest reported quarter, Q2 FY2027 (quarter ended 2026-07-26), printed a GAAP gross margin of [state]">75.0% on revenue of $96.2B [Q2 FY2027] — up from [Q1 FY2027]">74.9% in Q1 FY2027 (quarter ended 2026-04-26) [Q1 FY2027] and [Q2 FY2026]">72.4% a year earlier in Q2 FY2026 (quarter ended 2025-07-27) [Q2 FY2026]. But management guided Q3 FY2027 GAAP gross margin to [guidance]">74.0% ±50bp [guidance], and then a trough of 71–72% in Q4 FY2027 "on memory cost inflation" [earnings recap], recovering only to 72–73% for FY2028 [V, call]. The margin-slip trigger (set at 73%) is scheduled to fire on management's own numbers [watch]. In plain terms: memory scarcity is costing NVIDIA roughly 3–4 points of gross margin from the Q2 peak ([state]">75.0% reported vs 71–72% guided), and the question is whether it stays a one-quarter reset.
Management also stated FY2028 revenue growth of ~70% is supply-constrained [V, call] — so the ceiling on upside, not just margin, sits with suppliers.
Midstream — assembly (Hon Hai, Wistron: flowing, but leveraged)
The assembly layer looks healthy. Hon Hai (Foxconn), NVIDIA's largest assembly partner, reported record August sales, up 52%, with Europe lining up its own AI orders [supply-chain agent]. That is good on two counts: volumes are moving through the chain, and European demand modestly diversifies away from US hyperscalers. Wistron raised $1.47B via GDR to fund raw materials for its AI server ramp [supply-chain agent] — supportive of capacity, though the risk register notes the dilution and flags "supplier financial stress" [risk], meaning ODMs are stretching balance sheets to keep up.
One structural wrinkle: Hon Hai sits on both sides of NVIDIA's chain — a named supplier and a leading candidate for the large anonymized direct customers, since ODMs buy GPUs to build systems [supply-chain agent]. The reported customer-concentration figures therefore partly describe the same handful of companies twice.
Downstream — customers and financing (the real fault line)
Customer concentration is a high, live risk: "significant revenue stems from limited partners" [risk]. Demand itself is not in question — $96.2B reported and [guidance]">$108.0B ±[guidance]">2% guided for Q3 FY2027 [guidance]. The concern is how that demand is financed:
Balance-sheet fingerprints of a stretched chain
Three reported lines corroborate the picture. Receivables jumped to [Q2 FY2027]">$63.1B (DSO ~60 days) in Q2 FY2027 from [Q1 FY2027]">$40.7B (~45 days) in Q1 FY2027 [Q2 FY2027][Q1 FY2027] — a [Q2 FY2027]">$22.4B one-quarter build, and up from [Q2 FY2026]">$27.8B a year ago [Q2 FY2026]. Customers are being carried longer. Inventories doubled year over year to [Q2 FY2027]">$31.6B from [Q2 FY2026]">$15.0B [Q2 FY2027][Q2 FY2026], consistent with pre-buying scarce components. And total debt rose to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B after [Q2 FY2027]">$24.9B of issuance [Q2 FY2027], while non-marketable securities grew to $51.2B and a $2.9B payment went to Groq [F, quarterly/NVDA] — NVIDIA is itself becoming a financier of the ecosystem. Credit remains a non-issue (0 warning signals) [company analyst], but the direction is clear.
What to watch: (1) Whether Q4 FY2027 gross margin lands inside 71–72% and whether memory inflation extends into FY2028 [earnings recap]; (2) Samsung/SK Hynix allocation commitments to NVIDIA versus competitors [supply-chain agent]; (3) confirmation or denial of the OpenAI deal structure and CoreWeave CDS staying below [watch]">600bps [watch]; (4) whether the receivables build reverses next quarter.
Bottom line: The chain is delivering record volume, but memory scarcity has already cost ~3–4 points of guided margin, and the customers absorbing that volume are increasingly financed — partly by NVIDIA itself.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
What is new is memory. Samsung Electronics and SK Hynix inventories have dropped below 10 days, described as a critical shortage of HBM/DRAM (the high-bandwidth memory stacked next to every data-center GPU) that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" . A second report corroborates a "hard capacity ceiling near-term" for module output . The risk register now marks supply chain & manufacturing as LIVE elevated on this headline .
The short version: NVIDIA's chain is concentrated at every layer, and the layer under the most stress right now is memory – not foundry, not demand. NVIDIA is fabless: it designs the GPUs and depends on third-party foundries, a handful of memory makers, and Taiwanese ODMs to build the systems. That structure means a supply shock hits NVIDIA's cost of goods (via supplier pricing) and its ability to ship, rather than its own factories. The near-term evidence is exactly that pattern: memory suppliers are running out of inventory, and management has already guided gross margin down because of it.
What is new is memory. Samsung Electronics and SK Hynix inventories have dropped below 10 days, described as a critical shortage of HBM/DRAM (the high-bandwidth memory stacked next to every data-center GPU) that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" . A second report corroborates a "hard capacity ceiling near-term" for module output . The risk register now marks supply chain & manufacturing as LIVE elevated on this headline .
- OpenAI: a reported ~$250B financing/compute deal raised "circular financing" worries (NVIDIA capital funding its own demand) and drove a ~4% drop on 2026-07-27; the deal's structure – equity vs vendor financing – is unconfirmed . - Neoclouds: CoreWeave's credit-default-swap spread – the market's price for insuring its debt – has recovered from ~881bps to ~452bps; above 600bps would signal renewed stress . - Hyperscalers: a coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL is flagged as "THE highest-signal bear trigger" . - China: guidance excludes any data-center compute revenue from China ; Huawei's new high-end chip marks competition as LIVE elevated .
- OpenAI: a reported ~$250B financing/compute deal raised "circular financing" worries (NVIDIA capital funding its own demand) and drove a ~4% drop on 2026-07-27; the deal's structure – equity vs vendor financing – is unconfirmed . - Neoclouds: CoreWeave's credit-default-swap spread – the market's price for insuring its debt – has recovered from ~881bps to ~452bps; above 600bps would signal renewed stress . - Hyperscalers: a coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL is flagged as "THE highest-signal bear trigger" . - China: guidance excludes any data-center compute revenue from China ; Huawei's new high-end chip marks competition as LIVE elevated .
Midstream – assembly (Hon Hai, Wistron: flowing, but leveraged)
Caveat up front: I can't pull the filing live, so this is based on the most recent NVDA 10-Q I can speak to — quarter ended Oct 26, 2025 (Q3 FY26, filed Nov 19, 2025, same day as the call: $57.0B revenue, $51.2B data center, Q4 guide $65B ±[guidance]">2%). If a newer 10-Q is out, apply the same checklist to it — the pattern is remarkably stable.
What the 10-Q said that the call skipped or soft-pedaled
Customer concentration. Two direct customers were ~22% and ~[SEC filings]">15% of revenue (indirect end-customers even higher, one estimated in the low-teens or above). Management said "broad-based demand" on the call; the filing says roughly a third of revenue runs through two invoices. That's before you add the hyperscaler capex circularity (OpenAI/CoreWeave-style equity-for-compute arrangements, also disclosed in the filing's investments note).
Receivables outran revenue. AR rose to ~[Q3 FY2026]">$33B from ~[Q2 FY2026]">$28B, faster than the ~22% sequential revenue growth. DSO drifted into the low-50s. Not alarming for a supplier to Microsoft/Meta, but it's quietly building — cash conversion was ~[Q1 FY2026]">$22B FCF on [Q3 FY2026]">$32B operating income.
Inventory and purchase commitments. Inventory ~[Q3 FY2026]">$20B (up ~[state]">30% q/q); supply and capacity commitments ~$50B (up from ~$44B). Call framing: "supply visibility." 10-Q framing: contractual obligations to buy CoWoS, HBM, and substrate whether or not Rubin ramps on schedule. Same fact, different tail risk.
China: guidance excludes it, but so does the balance sheet. The 10-Q reiterated no H20 sales to China in the quarter, no license regime finalized for the [SEC filings]">15% revenue-share arrangement, and Blackwell unlicensed for China. Colette mentioned the exclusion; the 10-Q also notes prior H20 inventory reserves haven't been reversed — i.e., they don't expect it back.
Gross margin composition. GAAP GM [Q3 FY2026]">73.4%; Q4 guide ~[state]">75%. The 10-Q disclosed inventory provisions and releases netting out roughly neutral. Meaning the margin expansion to [state]">75% is meant to be clean mix/yield improvement, not reserve releases — which is good, but leaves no cushion if Blackwell Ultra/Rubin yields wobble.
Which of these should change how you read the guide
Bottom line: The call sells breadth and visibility; the 10-Q shows a business where ~37% of revenue is two customers, ~$50B of committed supply is riding on their capex plans, and receivables are stretching — so treat the $65B guide as high-probability but binary on hyperscaler capex, and watch AR and purchase commitments in the next filing as the real guidance-risk indicators.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
The short answer: the 10-Q filed 2026-08-26 carried at least four material items the call quotes on record don't touch — a $279B supply-commitment book, a [Q3 FY2026]">$36B guarantee to buy unsold AI-cloud capacity, $500B of third-party financing MOUs with optional residual-value support, and a fourfold jump in debt with a new indebtedness risk factor. The first two change how you should read the Q3 FY2027 (quarter ending late October 2026) revenue guide; the debt and receivables lines change how you should read the capital-return story. One honest aside: I have the call's key quotes, not a verbatim transcript, so "not said on the call" here means "absent from the commitments and guidance management put on record."
What the call did say
Management guided (not reported) Q3 FY2027 revenue of [guidance]">$108.0B ±[guidance]">2%, GAAP gross margin [guidance]">74.0% ±50bps, GAAP opex ~[guidance]">$9.2B, and explicitly excluded any China Data Center compute revenue [guidance]. It also confirmed Vera Rubin production shipments began in August with orders from every major hyperscaler, and that Q2 saw a record [Q2 FY2026]">$26B returned to shareholders [transcripts]. Full-year opex growth was raised to the "low 50s" — the third upward opex revision in two fiscal years [transcripts].
What the 10-Q added that the call didn't
1. Supply commitments more than doubled to $279B. Purchase obligations to suppliers went from $119B last quarter to $279B as of July 26, 2026 [SEC filings]. That is a reported figure, and it is roughly 2.9x the $96.2B of reported Q2 FY2027 (quarter ended July 26, 2026) revenue [Q2 FY2027]. It supports the demand story behind the [guidance]">$108B guide, but the filing itself pairs it with the warning that customers "may postpone purchasing new architectures" for lack of data-center infrastructure or capital [SEC filings]. Read alongside reported inventories rising to [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B in Q1 FY2027 [Q2 FY2027] [Q1 FY2027], this is the line that converts a demand miss into a margin problem.
2. A new business model with [Q3 FY2026]">$36B of capacity-purchase guarantees. NVIDIA now commits to buy unsold capacity from select AI-cloud partners; commitments are typically six years and totaled [Q3 FY2026]">$36B as of July 26, 2026 [SEC filings]. The risk-factor section restates it plainly: NVIDIA guarantees the purchases if capacity goes unsold to third parties [SEC filings]. This is contingent demand for NVIDIA's own product, and it lands on the same "circular financing" nerve as the reported ~$250B OpenAI deal that knocked the stock ~4% on 2026-07-27 [watch].
3. $500B of financing MOUs. In August NVIDIA signed memoranda to mobilize "more than $500 billion of third-party capital" for AI infrastructure; the funding is meant to be independently underwritten, but "at our option, we may provide limited residual-value support" [SEC filings]. That last clause means NVIDIA may backstop asset values on hardware it sold.
4. Debt up fourfold, plus a new indebtedness risk factor. Reported total debt rose to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B in Q1 FY2027, with [Q2 FY2027]">$24.9B issued fiscal year-to-date against [Q2 FY2027]">$39.0B of buybacks and [Q2 FY2027]">$6.3B of dividends [Q2 FY2027]. The 10-Q added a standalone risk factor on debt service and covenants [SEC filings]. The "record [Q2 FY2026]">$26B returned" line on the call was, in part, borrowed.
5. Receivables stretch. Reported receivables hit [Q2 FY2027]">$63.1B, DSO ~60 days, versus [Q1 FY2027]">$40.7B / ~45 days in Q1 FY2027 and ~54 days a year earlier in Q2 FY2026 [Q2 FY2027] [Q1 FY2027] [Q2 FY2026]. Revenue nearly doubled y/y; receivables more than doubled.
6. China: the exclusion is not conservatism, it's the fact pattern. The filing confirms the PRC government blocked H20 sales despite U.S. licenses and the inventory couldn't be sold [SEC filings]. New risk language adds that foreign governments may steer customers to competitors and that export rules trigger foreign inquiries [SEC filings]. Excluding China from guidance sounded prudent on the call; the 10-Q says there was nothing to include.
7. Smaller items. One direct customer at [guidance]">16% of Q2 revenue [SEC filings]; supply constraints acknowledged for both Blackwell and Rubin [SEC filings]; "land and shell" added to the infrastructure bottleneck list, and AI clouds noted as lacking long-term contracts [SEC filings]; marketable securities narrowed to U.S. government paper [SEC filings].
Which ones should change how you read the guidance
What to watch: whether the [Q3 FY2026]">$36B capacity commitment grows in the Q3 10-Q (late November) and whether any residual-value support under the $500B MOUs becomes a definitive agreement; DSO — another move above ~60 days alongside a further inventory build; total debt vs. buyback pace, and whether the $279B supply figure keeps compounding faster than revenue.
Bottom line: The call sold demand and capital return; the filing disclosed that NVIDIA is now partly financing both — read the [guidance]">$108B guide as well-supported but less independent, and the Q4 margin step-down as the first scheduled test of whether that support ever hits the P&L.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
1. Supply commitments more than doubled to $279B. Purchase obligations to suppliers went from $119B last quarter to $279B as of July 26, 2026 . That is a reported figure, and it is roughly 2.9x the $96.2B of reported Q2 FY2027 (quarter ended July 26, 2026) revenue Q2 FY2027. It supports the demand story behind the $108B guide, but the filing itself pairs it with the warning that customers "may postpone purchasing new architectures" for lack of data-center infrastructure or capital . Read alongside reported inventories rising to $31.6B from $25.8B in Q1 FY2027 Q2 FY2027 Q1 FY2027, this is the line that converts a demand miss into a margin problem.
The short answer: the 10-Q filed 2026-08-26 carried at least four material items the call quotes on record don't touch – a $279B supply-commitment book, a $36B guarantee to buy unsold AI-cloud capacity, $500B of third-party financing MOUs with optional residual-value support, and a fourfold jump in debt with a new indebtedness risk factor. The first two change how you should read the Q3 FY2027 (quarter ending late October 2026) revenue guide; the debt and receivables lines change how you should read the capital-return story. One honest aside: I have the call's key quotes, not a verbatim transcript, so "not said on the call" here means "absent from the commitments and guidance management put on record."
7. Smaller items. One direct customer at 16% of Q2 revenue ; supply constraints acknowledged for both Blackwell and Rubin ; "land and shell" added to the infrastructure bottleneck list, and AI clouds noted as lacking long-term contracts ; marketable securities narrowed to U.S. government paper .
6. China: the exclusion is not conservatism, it's the fact pattern. The filing confirms the PRC government blocked H20 sales despite U.S. licenses and the inventory couldn't be sold . New risk language adds that foreign governments may steer customers to competitors and that export rules trigger foreign inquiries . Excluding China from guidance sounded prudent on the call; the 10-Q says there was nothing to include.
4. Debt up fourfold, plus a new indebtedness risk factor. Reported total debt rose to $33.4B from $8.5B in Q1 FY2027, with $24.9B issued fiscal year-to-date against $39.0B of buybacks and $6.3B of dividends Q2 FY2027. The 10-Q added a standalone risk factor on debt service and covenants . The "record $26B returned" line on the call was, in part, borrowed.
Additional questions, model comparisons and research costs. Open what you want to inspect.
Four saved questions asked three ways on Claude Fable 5.1. Scoring v2 uses frozen evidence and contextual checklist matches. Original answers are unchanged; earlier accuracy rankings have been withdrawn.
QuestionWhat did the latest 10-Q disclose that management did not say on the earnings call the same day, and which of those disclosures should change how I read the guidance?
Rescored under proof-v2. The earlier overall ranking is withdrawn because its numeric and coverage checks were unreliable. Inspect the checklist passages and complete answers below; content coverage and recorded cost do not establish factual accuracy.
I'll pull the latest filing and call details.Short answer: the 10-Q's headline number is [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B of supply commitments—up from [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119B a quarter ago—plus roughly $250B of customer-side financing exposure ($105B guarantees, $99B private equity stakes, [Q2 FY2027]">$25B equity commitments, ~$29–36B cloud-capacity purchase obligations) that the call described qualitatively but never sized. Those numbers are what turn "supply-constrained 70% growth" from a demand story into a balance-sheet story.
Q3 revenue guided to [guidance]">$108B ±[guidance]">2%; Q3 gross margin [guidance]">74% ±50bp; Q4 gross margin expected to bottom at 71–72%; FY28 gross margin settling at 72–73%. Kress guided FY28 revenue growth of 70% vs. Street at 44%, saying customer forecasts "point to our growth doubling next year" but that guidance reflects supply constraints. Management also touted a new revenue-sharing model with NeoClouds and partnerships with infrastructure capital providers to raise over $500B, and record shareholder returns of [Q2 FY2026]">$26B with a commitment to return excess free cash flow.
1. Supply commitments: [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B. Supply and capacity commitments surged to [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B as of July 26, 2026, up from [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119B last quarter—a material increase in off-balance sheet obligations driven by long-term agreements for data center infrastructure, memory, and manufacturing facilities. Against $95.2B at January year-end, the book ran 2.34x in one quarter and 2.93x across six months; the company attributes the increase primarily to memory.
2. Capacity-buyback obligations to NeoClouds. NVIDIA is now entering agreements where it commits to purchase cloud capacity if AI cloud partners cannot sell it to third parties, exposing the company to potential underutilization risk. Cloud service agreement commitments increased to [10-Q notes]">$29B (vs. $13.7B in Q2 FY26). One independent read of the filing puts the residual-capacity commitment at [Q3 FY2026]">$36B to buy back cloud capacity—the two figures may be separate line items; check Note 10 directly.
3. The $105B SB Energy/OpenAI guarantees. Guarantees capped at $105B provide credit support for 4.25 GW of data center leases for OpenAI; these are 20-year guarantees with exposure declining as lease payments are made. Mitigants: OpenAI has agreed to reimburse and indemnify NVIDIA for any amounts paid, and each takes effect when its lease commences, from 2028, and pays only on tenant insolvency. The filed form also contains a covenant that the guaranteed premises must exclusively install and deploy NVIDIA reference architecture for at least 20 years. Caveat: utility service is contracted for only 0.8 of ~10 GW of design load, retail power agreements aren't executed, and either side can terminate if development milestones are missed.
4. Private-company equity: $99B. Equity investments in private companies and infrastructure financiers reached $99B (vs. [10-Q notes]">$22.3B at January 25); equity investment commitments totaled [Q2 FY2027]">$25B; equity-method stakes in infrastructure financiers were $3.3B with maximum loss exposure of $4.7B. NVIDIA determined it is not the primary beneficiary of its VIE investments and does not consolidate them.
5. Leverage. [Q2 FY2027]">$25B of new senior notes in June across seven tranches took total debt to [Q3 FY2026]">$33.4B (from [Q1 FY2026]">$8.5B in January); long-term debt rose to [10-Q notes]">$32.4B; CP capacity is [Q2 FY2027]">$25B with nothing drawn. Remaining buyback authorization is $99.3B.
6. Calendar. FY27 is a 53-week year; Q4 FY27 will be a 14-week quarter.
Bottom line: The call sold 70% supply-constrained growth; the 10-Q shows NVIDIA bought that supply with [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B of mostly-memory commitments and is underwriting a meaningful slice of the demand with ~$250B of guarantees, equity and capacity puts—so read the guide as high-conviction but with the downside now sitting in purchase obligations and counterparty credit rather than in order slippage.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
I'll pull the latest filing and call details.Short answer: the 10-Q's headline number is $279B of supply commitments–up from $119B a quarter ago–plus roughly $250B of customer-side financing exposure ($105B guarantees, $99B private equity stakes, $25B equity commitments, ~$29–36B cloud-capacity purchase obligations) that the call described qualitatively but never sized. Those numbers are what turn "supply-constrained 70% growth" from a demand story into a balance-sheet story.
I'll pull the latest filing and call details.Short answer: the 10-Q's headline number is $279B of supply commitments–up from $119B a quarter ago–plus roughly $250B of customer-side financing exposure ($105B guarantees, $99B private equity stakes, $25B equity commitments, ~$29–36B cloud-capacity purchase obligations) that the call described qualitatively but never sized. Those numbers are what turn "supply-constrained 70% growth" from a demand story into a balance-sheet story.
5. Leverage. $25B of new senior notes in June across seven tranches took total debt to $33.4B (from $8.5B in January); long-term debt rose to $32.4B; CP capacity is $25B with nothing drawn. Remaining buyback authorization is $99.3B.
I'll pull the latest filing and call details.Short answer: the 10-Q's headline number is $279B of supply commitments–up from $119B a quarter ago–plus roughly $250B of customer-side financing exposure ($105B guarantees, $99B private equity stakes, $25B equity commitments, ~$29–36B cloud-capacity purchase obligations) that the call described qualitatively but never sized. Those numbers are what turn "supply-constrained 70% growth" from a demand story into a balance-sheet story.
I'll pull the latest filing and call details.Short answer: the 10-Q's headline number is $279B of supply commitments–up from $119B a quarter ago–plus roughly $250B of customer-side financing exposure ($105B guarantees, $99B private equity stakes, $25B equity commitments, ~$29–36B cloud-capacity purchase obligations) that the call described qualitatively but never sized. Those numbers are what turn "supply-constrained 70% growth" from a demand story into a balance-sheet story.
Short answer: the 10-Q filed 2026-08-26 for Q2 FY2027 (quarter ended July 26, 2026) put on paper roughly $450B of new commitments and guarantees that the call excerpts I have never quantify — supply commitments more than doubled to [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B, a [Q3 FY2026]">$36B pledge to buy unsold AI-cloud capacity, a $105B guarantee for SB Energy, [Q2 FY2027]">$25B of new debt, and a $500B third-party financing platform. The call's quotable lines were about Rubin shipping, a record [Q2 FY2026]">$26B capital return, and OpEx growing in the "low 50s." The disclosures that should change how you read the [guidance]">$108B Q3 guide are the capacity-purchase commitment, the supply-commitment jump, and the receivables/customer-advance picture — because together they mean part of demand is now being financed or underwritten by NVIDIA itself, and the guided margin slide to 71–72% by Q4 is the cost of the supply build. (I can only compare against the call quotes on file; the full transcript may have touched some items.)
What the call said Three things surface in the transcript record: Vera Rubin production shipments began in August with purchase orders "from every major hyperscaler, AI cloud and system OEM"; a record [Q2 FY2026]">$26B returned to shareholders in Q2 ([Q3 FY2026]">$20B buybacks, [Q2 FY2027]">$6B dividends); and full-year OpEx growth raised to the "low 50s" — the second upward revision this year, after "low 40s" in February and "upper forties" in May [transcripts]. Guidance for Q3 FY2027 (guided, not reported): revenue [guidance]">$108.0B ±[guidance]">2%, GAAP and non-GAAP gross margin [guidance]">74.0% ±50bps, GAAP OpEx ~[guidance]">$9.2B, tax 16–18%, explicitly excluding any China Data Center compute revenue [guidance].
What the 10-Q added — and which items matter for the guide
1. Supply commitments [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119B → [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B (reported). Changes the margin read. The filing says supply and capacity commitments went from [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119B last quarter to [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B as of July 26, with [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$92B due in the rest of FY2027 and [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$87B in year one [10-Q notes]. Total future purchase and other commitments are [10-Q notes]">$366B [10-Q notes]. This is a $160B step-up in one quarter (279 minus 119) against a quarterly revenue run-rate of $96.2B (reported, Q2) [Q2 FY2027]. Inventory is already [Q2 FY2027]">$31.6B, up from [Q1 FY2027]">$25.8B in Q1 and [Q2 FY2026]">$15.0B a year ago [Q2 FY2027] [Q1 FY2027] [Q2 FY2026], and inventory provisions ran $784M in the quarter [10-Q notes]. The MD&A pairs the commitment with a warning that customers may postpone purchases for lack of data-center infrastructure or capital [SEC filings]. So the [guidance]">74% Q3 gross-margin guide — and the further slide to 71–72% guided for Q4 FY2027 — should be read as the Rubin/memory ramp cost, with the gm_slip trigger (below 73%) scheduled to fire on management's own outlook [watch]. Q2's [state]">75.0% reported margin [Q2 FY2027] is the peak of the cycle unless the mix improves.
2. New AI-cloud model: NVIDIA buys unsold capacity — [Q3 FY2026]">$36B (reported). Changes the revenue-quality read. The filing introduces a model where NVIDIA commits to purchase capacity from select AI-cloud partners if third parties don't; commitments are typically six years and total [Q3 FY2026]">$36B, with a matching new risk factor [SEC filings] [10-Q notes]. Add [10-Q notes]">$29B of cloud service agreements, [Q2 FY2027]">$25B of equity-investment commitments ([10-Q notes]">$18B due this fiscal year) and [Q2 FY2027]">$25B of data-center leases not yet commenced [10-Q notes]. Some slice of the GPUs in the [guidance]">$108B guide sell to clouds whose downside NVIDIA now backstops. There's no revenue-at-risk figure, but this is the "circular financing" concern the market flagged when the OpenAI deal report knocked the stock ~4% on 2026-07-27 [watch] — now in a filing, not a headline.
3. SB Energy guarantee $105B and $500B financing MOUs (reported). Changes the balance-sheet read. Maximum guarantee exposure is [10-Q notes]">$108.5B: $105B to affiliates of SB Energy for land, power and shell buildout (entered August 2026), plus [10-Q notes]">$3.5B for AI-cloud land/power/shell [10-Q notes]. Separately, MOUs to mobilize >$500B of third-party capital, where NVIDIA "may provide limited residual-value support" [SEC filings]. Financing is underwritten by others, but the residual-value line means NVIDIA can end up holding the tail if hardware values fall faster than expected.
4. Debt [Q1 FY2026]">$8.5B → [Q3 FY2026]">$33.4B (reported). [10-Q notes]">$25.0B of senior unsecured notes across seven tranches issued in June 2026 [10-Q notes], and a brand-new indebtedness risk factor [SEC filings]. Cash rose to [Q2 FY2027]">$22.4B, but the company borrowed while returning $45.3B YTD ([Q2 FY2027]">$39.0B buybacks + [Q2 FY2027]">$6.3B dividends) [Q2 FY2027], with $99.3B of buyback authorization remaining [10-Q notes].
5. Receivables and customer advances — two-sided. Receivables jumped to [Q2 FY2027]">$63.1B, DSO ~60 days, from [Q1 FY2027]">$40.7B/~45 days in Q1 [Q2 FY2027] [Q1 FY2027]: customers are paying more slowly as volumes surge. Offsetting it, deferred-revenue additions were [10-Q notes]">$17.7B in the first half including [10-Q notes]">$15.6B of customer advances, versus [10-Q notes]">$8.3B/[10-Q notes]">$7.5B a year earlier [10-Q notes] — genuine prepaid demand that supports the Q3 guide. Five customers hold 22/14/13/11/[10-Q notes]">10% of receivables and one customer was [guidance]">16% of Q2 revenue [10-Q notes].
6. China: the guide excludes it, and the filing explains why. The PRC government blocked H20 sales despite U.S. licenses, so NVIDIA couldn't move that inventory [SEC filings]; the prior China antitrust disclosure was folded into general export-control language, and new risks note foreign governments may steer customers to competitors [SEC filings]. The "conservative" China exclusion is a floor, not upside.
7. Investments. Non-marketable securities rose to [10-Q notes]">$51.2B from [10-Q notes]">$22.3B; [10-Q notes]">$36.9B of public equity is under lock-up; first-half unrealized gains were $12.5B [10-Q notes] — earnings quality is now partly mark-to-market on partner stakes.
Also disclosed, not material to this question: no accrued legal contingencies; marketable-securities description narrowed to U.S. government debt; export controls linked to foreign inquiries; land/shell added to infrastructure constraints; Data Center revenue $89.0B (+117% y/y) [10-Q notes] [SEC filings].
Bottom line: The call sold the demand; the filing showed NVIDIA now finances, guarantees and pre-buys a growing share of it — read the [guidance]">$108B guide as achievable but increasingly self-supported, and the 71–72% Q4 margin as the price.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
1. Supply commitments $119B → $279B (reported). Changes the margin read. The filing says supply and capacity commitments went from $119B last quarter to $279B as of July 26, with $92B due in the rest of FY2027 and $87B in year one . Total future purchase and other commitments are $366B . This is a $160B step-up in one quarter (279 minus 119) against a quarterly revenue run-rate of $96.2B (reported, Q2) Q2 FY2027. Inventory is already $31.6B, up from $25.8B in Q1 and $15.0B a year ago Q2 FY2027 Q1 FY2027 Q2 FY2026, and inventory provisions ran $784M in the quarter . The MD&A pairs the commitment with a warning that customers may postpone purchases for lack of data-center infrastructure or capital . So the 74% Q3 gross-margin guide – and the further slide to 71–72% guided for Q4 FY2027 – should be read as the Rubin/memory ramp cost, with the gmslip trigger (below 73%) scheduled to fire on management's own outlook . Q2's 75% reported margin Q2 FY2027 is the peak of the cycle unless the mix improves.
Short answer: the 10-Q filed 2026-08-26 for Q2 FY2027 (quarter ended July 26, 2026) put on paper roughly $450B of new commitments and guarantees that the call excerpts I have never quantify – supply commitments more than doubled to $279B, a $36B pledge to buy unsold AI-cloud capacity, a $105B guarantee for SB Energy, $25B of new debt, and a $500B third-party financing platform. The call's quotable lines were about Rubin shipping, a record $26B capital return, and OpEx growing in the "low 50s." The disclosures that should change how you read the $108B Q3 guide are the capacity-purchase commitment, the supply-commitment jump, and the receivables/customer-advance picture – because together they mean part of demand is now being financed or underwritten by NVIDIA itself, and the guided margin slide to 71–72% by Q4 is the cost of the supply build. (I can only compare against the call quotes on file; the full transcript may have touched some items.)
5. Receivables and customer advances – two-sided. Receivables jumped to $63.1B, DSO ~60 days, from $40.7B/~45 days in Q1 Q2 FY2027 Q1 FY2027: customers are paying more slowly as volumes surge. Offsetting it, deferred-revenue additions were $17.7B in the first half including $15.6B of customer advances, versus $8.3B/$7.5B a year earlier – genuine prepaid demand that supports the Q3 guide. Five customers hold 22/14/13/11/10% of receivables and one customer was 16% of Q2 revenue .
6. China: the guide excludes it, and the filing explains why. The PRC government blocked H20 sales despite U.S. licenses, so NVIDIA couldn't move that inventory ; the prior China antitrust disclosure was folded into general export-control language, and new risks note foreign governments may steer customers to competitors . The "conservative" China exclusion is a floor, not upside.
Short answer: the 10-Q filed 2026-08-26 for Q2 FY2027 (quarter ended July 26, 2026) put on paper roughly $450B of new commitments and guarantees that the call excerpts I have never quantify – supply commitments more than doubled to $279B, a $36B pledge to buy unsold AI-cloud capacity, a $105B guarantee for SB Energy, $25B of new debt, and a $500B third-party financing platform. The call's quotable lines were about Rubin shipping, a record $26B capital return, and OpEx growing in the "low 50s." The disclosures that should change how you read the $108B Q3 guide are the capacity-purchase commitment, the supply-commitment jump, and the receivables/customer-advance picture – because together they mean part of demand is now being financed or underwritten by NVIDIA itself, and the guided margin slide to 71–72% by Q4 is the cost of the supply build. (I can only compare against the call quotes on file; the full transcript may have touched some items.)
Short answer: the 10-Q filed 2026-08-26 for Q2 FY2027 (quarter ended July 26, 2026) put on paper roughly $450B of new commitments and guarantees that the call excerpts I have never quantify – supply commitments more than doubled to $279B, a $36B pledge to buy unsold AI-cloud capacity, a $105B guarantee for SB Energy, $25B of new debt, and a $500B third-party financing platform. The call's quotable lines were about Rubin shipping, a record $26B capital return, and OpEx growing in the "low 50s." The disclosures that should change how you read the $108B Q3 guide are the capacity-purchase commitment, the supply-commitment jump, and the receivables/customer-advance picture – because together they mean part of demand is now being financed or underwritten by NVIDIA itself, and the guided margin slide to 71–72% by Q4 is the cost of the supply build. (I can only compare against the call quotes on file; the full transcript may have touched some items.)
5. Receivables and customer advances – two-sided. Receivables jumped to $63.1B, DSO ~60 days, from $40.7B/~45 days in Q1 Q2 FY2027 Q1 FY2027: customers are paying more slowly as volumes surge. Offsetting it, deferred-revenue additions were $17.7B in the first half including $15.6B of customer advances, versus $8.3B/$7.5B a year earlier – genuine prepaid demand that supports the Q3 guide. Five customers hold 22/14/13/11/10% of receivables and one customer was 16% of Q2 revenue .
7. Investments. Non-marketable securities rose to $51.2B from $22.3B; $36.9B of public equity is under lock-up; first-half unrealized gains were $12.5B – earnings quality is now partly mark-to-market on partner stakes.
2. New AI-cloud model: NVIDIA buys unsold capacity – $36B (reported). Changes the revenue-quality read. The filing introduces a model where NVIDIA commits to purchase capacity from select AI-cloud partners if third parties don't; commitments are typically six years and total $36B, with a matching new risk factor . Add $29B of cloud service agreements, $25B of equity-investment commitments ($18B due this fiscal year) and $25B of data-center leases not yet commenced . Some slice of the GPUs in the $108B guide sell to clouds whose downside NVIDIA now backstops. There's no revenue-at-risk figure, but this is the "circular financing" concern the market flagged when the OpenAI deal report knocked the stock ~4% on 2026-07-27 – now in a filing, not a headline.
Short answer: The call sold the demand story (FY2028 revenue +~70%, "supply-constrained") and got ahead of the margin trough ([guidance]">74% → 71–72% → 72–73%). The 10-Q filed the same afternoon carries the balance-sheet and legal fine print behind that story — a receivables stretch, contingent purchase and guarantee obligations, a China door that is shut rather than merely "excluded," and equity-gain-inflated earnings. Five of those disclosures should change how you weight the Q3 guide of [guidance]">$108B ±[guidance]">2% at [guidance]">74% ±50bp gross margin, [guidance]">$9.2B GAAP opex [guidance].
What was on the call vs. only in the filing
Call coverage confirms management did address the big headline items: Kress said Nvidia expects fiscal 2028 revenue growth of 70%, while analysts were expecting 44% [web]; Kress said NVIDIA has invested nearly $50 billion in frontier AI laboratories and has partnered with Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR on financing platforms expected to raise more than $500 billion in third-party capital [web]; the [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B supply commitment was in the CFO commentary and Kress fielded a Q&A on it [web]; and the SB Energy/Ohio 4.25 GW site was mentioned [web]. So those are not "hidden." What the call did not say — or said only qualitatively — is below.
1. Receivables ballooned faster than revenue. Receivables went from [Q1 FY2027]">$40.7B (DSO ~45d) in Q1 FY2027 to [Q2 FY2027]">$63.1B (DSO ~60d) in Q2 FY2027 [Q1 FY2027][Q2 FY2027] — up 55% q/q against revenue up [guidance]">18% q/q ($81.6B → $96.2B). Five direct customers hold [10-Q notes]">22%, [10-Q notes]">14%, [10-Q notes]">13%, [state]">11% and [10-Q notes]">10% of that balance — ~70% [10-Q notes]. Nothing in the call flagged the collection lag. Paired with an accrual ratio of [state]">10.9% and a Piotroski F-score of 4 [state] (a 0–9 earnings-quality checklist; 4 is middling), this is the single most important quality signal for the guide: revenue is being recognized ahead of cash, and it is concentrated.
2. The AI-cloud "take-or-pay" model is a [Q3 FY2026]">$36B contingent liability, not just a revenue-share upside. On the call it was framed as NVIDIA can provide take-or-pay commitments on part of a facility's capacity to support project financing, then share in revenue generated above a defined minimum level. Kress said the model could create a recurring, usage-linked revenue stream [web]. The 10-Q says the commitments are typically six years, totaled [Q3 FY2026]">$36B as of 2026-07-26, and decrease as capacity is used by third-party customers "or by us for our research and development efforts" [SEC filings]; a new risk factor states NVIDIA guarantees the capacity purchases if unsold [SEC filings]. Read-through to guidance: if neocloud demand underperforms, unsold capacity becomes NVIDIA's own opex/R&D compute — the very line that has already forced two upward revisions to FY2027 opex growth (low 40s → upper 40s → low 50s, the latest citing "further increase in the usage of AI tools") [transcripts]. Also note the filing adds that AI clouds lack long-term contracts and that land and shell are now constraints [SEC filings].
3. [10-Q notes]">$108.5B of guarantees and [Q2 FY2027]">$25B of not-yet-commenced leases. Maximum guarantee exposure is [10-Q notes]">$108.5B, of which $105B is the August SB Energy land/power/shell credit support, plus [10-Q notes]">$3.5B for other AI clouds [10-Q notes]. Web reporting frames the $105B as residual-value guaranties on 20-year OpenAI leases, effective in phases from 2028 [web], and it followed reports of a $250B backstop under discussion [web] — smaller than the rumor, but it confirms the vendor-credit-support structure the openai_circular_financing trigger is watching [watch]. The $500B platforms are, per the 10-Q, MOUs that "may not lead to definitive agreements," with optional "limited residual-value support" from NVIDIA [SEC filings] — softer than the call's framing.
4. China is closed, not "conservatively excluded." The Q3 guide excludes China data center compute [guidance]. The 10-Q adds that PRC restrictions prevented H20 sales despite U.S. licenses, and that foreign governments may direct customers to competitors over export controls; the specific Mellanox/antitrust language was replaced by generalized text [SEC filings]. Net: the china_policy "optionality" [watch] is smaller than the call implied — no hidden upside cushion in the [guidance]">$108B.
5. Earnings and cash flow quality. H1 net income includes $12.5B of unrealized gains on public equities, only $1.5B of it in Q2 [10-Q notes] — strip this when extrapolating the Q3 guide to EPS. Non-marketable securities rose to [10-Q notes]">$51.2B from [10-Q notes]">$22.3B, [10-Q notes]">$36.9B of public holdings are lock-up restricted, and [10-Q notes]">$18B of equity-investment commitments fall in the rest of FY2027 [10-Q notes]. Meanwhile debt jumped to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B on a [Q2 FY2027]">$25B June note issue [10-Q notes][Q2 FY2027] with a new indebtedness risk factor [SEC filings], funding [Q2 FY2027]">$39B of YTD buybacks and [Q2 FY2027]">$6.3B dividends [Q2 FY2027].
How it re-frames the guidance
What to watch: Q3 receivables/DSO versus revenue growth (a DSO above ~60d again would confirm a trend), any drawdown of the [Q3 FY2026]">$36B AI-cloud commitment into NVIDIA's own usage, whether the $500B MOUs convert to definitive agreements, and CoreWeave CDS back above [watch]">600bps [watch].
Bottom line: The call's [guidance]">$108B/[guidance]">74% guide is credible on demand, but the 10-Q shows NVIDIA is increasingly financing that demand — via receivables, take-or-pay capacity, and [10-Q notes]">$108.5B of guarantees — so read the guide as a supply-constrained ceiling on revenue with rising, not falling, risk in margin, opex and cash conversion.
Sources (most reliable first): Nvidia (NVDA) Q2 2027 earnings report: Live updates [src:press] · NVIDIA Q2 Earnings Call Highlights [src:unverified]
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
2. The AI-cloud "take-or-pay" model is a $36B contingent liability, not just a revenue-share upside. On the call it was framed as NVIDIA can provide take-or-pay commitments on part of a facility's capacity to support project financing, then share in revenue generated above a defined minimum level. Kress said the model could create a recurring, usage-linked revenue stream . The 10-Q says the commitments are typically six years, totaled $36B as of 2026-07-26, and decrease as capacity is used by third-party customers "or by us for our research and development efforts" ; a new risk factor states NVIDIA guarantees the capacity purchases if unsold . Read-through to guidance: if neocloud demand underperforms, unsold capacity becomes NVIDIA's own opex/R&D compute – the very line that has already forced two upward revisions to FY2027 opex growth (low 40s → upper 40s → low 50s, the latest citing "further increase in the usage of AI tools") . Also note the filing adds that AI clouds lack long-term contracts and that land and shell are now constraints .
3. $108.5B of guarantees and $25B of not-yet-commenced leases. Maximum guarantee exposure is $108.5B, of which $105B is the August SB Energy land/power/shell credit support, plus $3.5B for other AI clouds . Web reporting frames the $105B as residual-value guaranties on 20-year OpenAI leases, effective in phases from 2028 , and it followed reports of a $250B backstop under discussion – smaller than the rumor, but it confirms the vendor-credit-support structure the openaicircularfinancing trigger is watching . The $500B platforms are, per the 10-Q, MOUs that "may not lead to definitive agreements," with optional "limited residual-value support" from NVIDIA – softer than the call's framing.
4. China is closed, not "conservatively excluded." The Q3 guide excludes China data center compute . The 10-Q adds that PRC restrictions prevented H20 sales despite U.S. licenses, and that foreign governments may direct customers to competitors over export controls; the specific Mellanox/antitrust language was replaced by generalized text . Net: the chinapolicy "optionality" is smaller than the call implied – no hidden upside cushion in the $108B.
5. Earnings and cash flow quality. H1 net income includes $12.5B of unrealized gains on public equities, only $1.5B of it in Q2 – strip this when extrapolating the Q3 guide to EPS. Non-marketable securities rose to $51.2B from $22.3B, $36.9B of public holdings are lock-up restricted, and $18B of equity-investment commitments fall in the rest of FY2027 . Meanwhile debt jumped to $33.4B from $8.5B on a $25B June note issue Q2 FY2027 with a new indebtedness risk factor , funding $39B of YTD buybacks and $6.3B dividends Q2 FY2027.
3. $108.5B of guarantees and $25B of not-yet-commenced leases. Maximum guarantee exposure is $108.5B, of which $105B is the August SB Energy land/power/shell credit support, plus $3.5B for other AI clouds . Web reporting frames the $105B as residual-value guaranties on 20-year OpenAI leases, effective in phases from 2028 , and it followed reports of a $250B backstop under discussion – smaller than the rumor, but it confirms the vendor-credit-support structure the openaicircularfinancing trigger is watching . The $500B platforms are, per the 10-Q, MOUs that "may not lead to definitive agreements," with optional "limited residual-value support" from NVIDIA – softer than the call's framing.
- The revenue guide is well-supported near-term: customer advances doubled to $15.6B , and Vera Rubin production shipments started in August with orders from every hyperscaler . But the growth is partly NVIDIA-underwritten (receivables, take-or-pay, guarantees). - The margin guide is a locked-in cost decision: $92B of supply commitments fall in the rest of FY2027 alone – roughly 1.6x the ~$56B of COGS implied by two quarters of ~$108B at 74% . Memory pricing is locked high; the filing notes agreements "may be cancelable... may result in additional costs" , so the Q4 71–72% path is scheduled to trip the gmslip trigger on management's own numbers . - Opex ($9.2B GAAP) has the least protection: the $36B capacity commitment is a direct feed into it if neocloud utilization disappoints.
5. Earnings and cash flow quality. H1 net income includes $12.5B of unrealized gains on public equities, only $1.5B of it in Q2 – strip this when extrapolating the Q3 guide to EPS. Non-marketable securities rose to $51.2B from $22.3B, $36.9B of public holdings are lock-up restricted, and $18B of equity-investment commitments fall in the rest of FY2027 . Meanwhile debt jumped to $33.4B from $8.5B on a $25B June note issue Q2 FY2027 with a new indebtedness risk factor , funding $39B of YTD buybacks and $6.3B dividends Q2 FY2027.
5. Earnings and cash flow quality. H1 net income includes $12.5B of unrealized gains on public equities, only $1.5B of it in Q2 – strip this when extrapolating the Q3 guide to EPS. Non-marketable securities rose to $51.2B from $22.3B, $36.9B of public holdings are lock-up restricted, and $18B of equity-investment commitments fall in the rest of FY2027 . Meanwhile debt jumped to $33.4B from $8.5B on a $25B June note issue Q2 FY2027 with a new indebtedness risk factor , funding $39B of YTD buybacks and $6.3B dividends Q2 FY2027.
QuestionWhat did the latest 10-Q disclose that management did not say on the earnings call the same day, and which of those disclosures should change how I read the guidance?
Rescored under proof-v2. The earlier overall ranking is withdrawn because its numeric and coverage checks were unreliable. Inspect the checklist passages and complete answers below; content coverage and recorded cost do not establish factual accuracy.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~[Q3 FY2025]">$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the [10-Q notes]">$253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / [guidance]">$4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
What the call said Q4 guide: revenue of $10.25 billion, plus or minus [Q2 FY2024]">$500 million, up 51% year over year; non-GAAP EPS of [guidance]">$4.02, plus or minus [guidance]">$0.20. Segments: Semiconductor Systems around $7.9 billion up 62% year over year, AGS about $1.84 billion up 22%, and other around $510 million. Margin: non-GAAP gross margin approximately 50.4% in Q4, up 32 basis points year over year. Demand framing: customers have found new ways to address cleanroom space constraints and significantly increased their demand for tool deliveries. Stock still fell — to $506.51, down 5.24%, as investors appeared to focus on the company's near-term margin outlook and the pace of future growth.
What the 10-Q disclosed that the call didn't
Long-dated backlog is small. Remaining unsatisfied performance obligations on contracts with an original estimated duration of one year or more was approximately [Q3 FY2025]">$1.0 billion, of which approximately 73% is expected to be recognized within 12 months, and the company excludes unsatisfied performance obligations for contracts with an original expected duration of one year or less. Translation: ~$730M of contractually visible long-dated revenue against a ~$10.25B quarterly guide. The guide is essentially a short-cycle book-and-ship number.
China quantified. China remained a major market at 28% of nine-month revenue ($6.69 billion); for the quarter, China contributed $2.51 billion in Q3, representing 28% of total sales. A year ago, that figure was 35%. So the [state]">25% headline growth was achieved with China roughly flat in dollars and shrinking in mix — non-China grew far faster than the headline.
BIS compliance tail. The filings reiterate that the company agreed to pay BIS [10-Q notes]">$253 million, paid in full in fiscal Q2, and the settlement agreement requires internal audits of export controls compliance and maintenance of export compliance training and reporting mechanisms. Not a P&L item going forward, but it constrains how aggressively AMAT can chase China mature-node share while rivals aren't similarly bound.
Margin attribution. MD&A credits the gain to higher volumes, increased average selling prices and lower material and manufacturing costs. Balance sheet/liquidity: operating cash flow was $5.57 billion (nine months), and cash, cash equivalents and investments totaled $14.5 billion; $100 million of commercial paper notes outstanding — immaterial.
Which of these should change how you read the guidance
The RPO number is the big one. With only ~[Q3 FY2025]">$1.0B of >1-year obligations and 73% of that landing in 12 months, the Q4 step-up from $9.12B to $10.25B (+12% q/q) rests on POs booked and shipped inside the quarter. That makes the ±[Q2 FY2024]">$500M range more live than usual — pushouts from a single large DRAM or foundry customer can swing it. It also means the call's "significantly increased demand for tool deliveries" is a near-term pull, not a multi-year contracted commitment.
The margin guide is flatter than the P&L momentum implies. Q3 non-GAAP GM was 50.4%; Q4 guide is 50.4% on [Q3 FY2026]">$1.1B more revenue. Given the 10-Q's ASP/material-cost attribution, flat GM at higher volume signals mix headwinds — China's shrinking share (historically higher-margin) and a heavier DRAM/advanced-packaging mix offsetting operating leverage. Read 50.4% as a ceiling, not a floor, for the next couple of quarters.
China at 28% is now the base, not a shrinking risk. Management's expectation that US export controls carve roughly $600 million out of fiscal 2026 revenue is already in the numbers; the 10-Q confirms China dollars are stabilizing at ~[Q3 FY2026]">$2.5B/quarter. Upside from here is non-China; further downside requires new rules.
Smaller items to keep in mind (from the release/call, not new in the 10-Q)
Bottom line: The 10-Q shows Q4's $10.25B guide is backed by only ~[Q3 FY2025]">$1.0B of long-dated contracted revenue and a margin outlook that's flat despite volume leverage — treat the guide as a high-confidence but short-cycle number with a genuine ±[Q2 FY2024]">$500M band, and treat 50.4% GM as the peak-mix ceiling, not the start of another leg up.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the $253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / $4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the $253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / $4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the $253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / $4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the $253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / $4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
I'll pull the latest filing and call details.Short answer: The Q3 FY26 10-Q (period ended July 26, 2026, filed with the Aug. 13 release) adds four things the call glossed over: (1) a thin long-dated backlog (~$1.0B RPO), (2) a hard China number (28% of revenue, ~$6.69B YTD), (3) ongoing BIS compliance obligations from the $253M settlement, and (4) MD&A attribution of margin gains to ASPs and material costs. Items 1 and 4 are the ones that should change how you read the $10.25B / $4.02 Q4 guide. Caveat up front: I could verify only parts of the 10-Q text, so check inventory, contract liabilities, and customer-concentration tables yourself.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message — "raised, record guide, supply-constrained" — didn't quantify: a [10-Q notes]">$698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while [Q1 FY2025]">$1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
What the call said (the bar) Management guided Q4 FY2026 (quarter ending late October 2026) revenue to $10,250M ± [Q2 FY2024]">$500M and non-GAAP EPS to [guidance]">$4.02 ± [guidance]">$0.20 (guided) [guidance], raised the calendar-2026 semi-systems growth forecast to >[state]">30%, and pointed to "unprecedented 8-quarter rolling customer forecasts"; the stock still fell 5.1% on supply-constraint worries [earnings recap agent]. Q3 FY2026 (quarter ended 2026-07-26) itself was reported revenue $9.1B ([state]">+25% y/y), GAAP gross margin [state]">50.3% [Q3 FY2026].
Disclosures that change how to read the guide
1. Prepayments are surging — the forward book is turning up. Contract liabilities (cash customers have paid ahead of delivery) stood at a reported $3.3B, up [10-Q notes]">$698M in the nine months to July 2026, versus a $379M decline in the year-ago nine months [10-Q notes]. AMAT recognized [Q3 FY2025]">$1.8B of revenue from the opening balance during the period [10-Q notes]. This is the "forward book re-accelerating" tell on the watch list, which had been flat at ~[watch]">1.1 months [watch]. Customers putting down money ahead of tools is the hardest evidence you can get that the $10.25B guide is demand-backed rather than hope-backed. Management referenced rolling forecasts; the filing shows cash.
2. Receivables are ballooning, and part of it is billing mechanics. Reported receivables were [Q3 FY2026]">$7.7B (DSO ~77 days), up from [Q2 FY2026]">$6.4B (~73d) in Q2 FY2026 (ended 2026-04-26) and [Q3 FY2025]">$5.8B (~72d) a year earlier [Q3 FY2026] [Q2 FY2026] [Q3 FY2025]. The filing attributes the rise to "unrecognized down payment invoices and reduced receivable factoring" [SEC filings]. Read that with #1: AMAT is invoicing down payments that haven't yet hit revenue (so they sit in receivables and contract liabilities), and it has stopped selling as many receivables for early cash. Nine-month operating cash flow was $5.6B, helped by higher net income and lower tax payments but "partially offset by a higher accounts receivable balance and higher vendor payments" [SEC filings]. The guide is a revenue number; the cash behind it is arriving later than the P&L suggests.
3. Inventory keeps building for a ramp that's supply-constrained. Inventories rose to a reported [Q2 FY2024]">$6.6B from [Q1 FY2025]">$6.3B in Q2 and [Q3 FY2025]">$5.8B a year ago [Q3 FY2026] [Q2 FY2026] [Q3 FY2025]; service spares alone are [10-Q notes]">$1.9B, up from [Q3 FY2025]">$1.8B at fiscal year-end [10-Q notes]. Against a guide implying roughly [Q3 FY2026]">$1.1B of sequential revenue growth ($10.25B guided vs $9.1B reported), that build is consistent with shipping into Q4 — but "higher vendor payments" means AMAT is paying suppliers ahead of collecting.
4. Capital return has been throttled. Buybacks were [Q1 FY2026]">$400M in Q3 and [Q3 FY2026]">$1.1B for nine months, versus [Q3 FY2025]">$4.0B in the year-ago nine months, with [10-Q notes]">$12.8B still authorized [10-Q notes]. Meanwhile the [Q1 FY2025]">$1.2B 3.300% notes due April 2027 moved into current maturities, taking noncurrent long-term debt to [10-Q notes]">$5.2B from [Q2 FY2026]">$6.5B at fiscal year-end [10-Q notes]. Cash and investments total a reported $14.5B at fair value (cost $13.4B, [Q1 FY2025]">$1.2B unrealized gains) [10-Q notes], with a [Q3 FY2025]">$4.0B commercial paper program available [10-Q notes]. AMAT can fund working capital, the 2027 maturity, and buybacks — but the slowdown says management is prioritizing the ramp over the share count while the guide is being delivered.
5. The tax line is running below what the EPS guide might assume. Q3 effective tax rate was [SEC filings]">12.7% ([SEC filings]">12.9% YTD), with a valuation allowance on CAMT credits cited as the main driver [SEC filings]. The [guidance]">$4.02 non-GAAP EPS guide explicitly excludes uncertain tax items [guidance]; the GAAP-to-non-GAAP gap can move if that allowance shifts.
6. Margin drivers are now cost, not just volume. The 10-Q added "lower material and manufacturing costs," "increases in average selling prices" and favorable mix as drivers of Semiconductor Systems margin [SEC filings]. That is why gross margin reached 50.3% — 3.4 points above the [watch]">46.9% watch threshold [watch] — and it argues the margin holds into a $10.25B quarter unless supply costs bite.
Disclosures that frame the China/geography risk Asia Pacific was [10-Q notes]">80% of Q3 revenue ([10-Q notes]">89% a year ago); foundry/logic [10-Q notes]">67% ([10-Q notes]">69%) [10-Q notes]; non-US revenue [10-Q notes]">85%, down from 88% [SEC filings]. The [10-Q notes]">$253M BIS settlement over China shipments and export-control compliance was expensed in the nine months and paid in Q2, with zero charge in Q3 [10-Q notes] — closed, but a reminder that the China watch item remains unquantified in the filing [watch].
Also disclosed, not material to this question: parent guarantees to banks of [10-Q notes]">$293M and maximum guarantee exposure of [10-Q n">$282M; restructuring reserve down to $12M from [10-Q notes]">$165M with [10-Q notes]">$12M of nine-month charges; [10-Q notes]">$7.5B short- and long-term investments ([10-Q notes]">$10.4B AFS debt securities), [Q2 FY2025]">$3.0B with no single maturity date, [10-Q notes]">$57M restricted cash; senior notes fair value [Estimate Of Fair Value Fair Value Disclosure, Senior Notes, Fair Value Inputs Level2] — “As of July 26, 2026, the aggregate principal amount of long-term senior unsecured notes i">$4.7B vs [Estimate Of Fair Value Fair Value Disclosure, Senior Notes, Fair Value Inputs Level2] — “As of July 26, 2026, the aggregate princi">$5.3B principal; immaterial private-equity impairments; and new FASB standards on income-tax disclosure, government grants and environmental credits [10-Q notes] [SEC filings].
What to watch: contract liabilities vs $3.3B and DSO vs 77 days in the Q4 FY2026 10-K; Q4 revenue vs the $9.75B–$10.75B guided range; whether the buyback re-accelerates from [Q1 FY2026]">$400M/quarter.
Bottom line: The filing strengthens the demand case behind the $10.25B guide (prepayments up [10-Q notes]">$698M) but shows the ramp is consuming cash — receivables, inventory and vendor payments up, buybacks down — so judge Q4 on cash conversion, not just the revenue print.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message – "raised, record guide, supply-constrained" – didn't quantify: a $698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while $1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message – "raised, record guide, supply-constrained" – didn't quantify: a $698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while $1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message – "raised, record guide, supply-constrained" – didn't quantify: a $698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while $1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
4. Capital return has been throttled. Buybacks were $400M in Q3 and $1.1B for nine months, versus $4.0B in the year-ago nine months, with $12.8B still authorized . Meanwhile the $1.2B 3.300% notes due April 2027 moved into current maturities, taking noncurrent long-term debt to $5.2B from $6.5B at fiscal year-end . Cash and investments total a reported $14.5B at fair value (cost $13.4B, $1.2B unrealized gains) , with a $4.0B commercial paper program available . AMAT can fund working capital, the 2027 maturity, and buybacks – but the slowdown says management is prioritizing the ramp over the share count while the guide is being delivered.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message – "raised, record guide, supply-constrained" – didn't quantify: a $698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while $1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
Disclosures that frame the China/geography risk Asia Pacific was 80% of Q3 revenue (89% a year ago); foundry/logic 67% (69%) ; non-US revenue 85%, down from 88% . The $253M BIS settlement over China shipments and export-control compliance was expensed in the nine months and paid in Q2, with zero charge in Q3 – closed, but a reminder that the China watch item remains unquantified in the filing .
6. Margin drivers are now cost, not just volume. The 10-Q added "lower material and manufacturing costs," "increases in average selling prices" and favorable mix as drivers of Semiconductor Systems margin . That is why gross margin reached 50.3–3.4 points above the 46.9% watch threshold – and it argues the margin holds into a $10.25B quarter unless supply costs bite.
5. The tax line is running below what the EPS guide might assume. Q3 effective tax rate was 12.7% (12.9% YTD), with a valuation allowance on CAMT credits cited as the main driver . The $4.02 non-GAAP EPS guide explicitly excludes uncertain tax items ; the GAAP-to-non-GAAP gap can move if that allowance shifts.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) put three things on the record that the call's headline message – "raised, record guide, supply-constrained" – didn't quantify: a $698M nine-month build in customer prepayments, a receivables jump to 77 days' sales driven by unbilled down payments and less factoring, and a buyback that has slowed to a quarter of last year's pace while $1.2B of debt moved to current. The first supports the $10.25B Q4 guide; the second and third tell you how much cash that guide is tying up.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) mostly confirms the call, but four things in the notes either weren't said on the call or were framed differently — and two of them genuinely change how to read the $10.25B ± [Q2 FY2024]">$500M / [guidance]">$4.02 ± [guidance]">$0.20 Q4 guide [guidance].
What the call said. Management's message was demand-led: record revenue of $9.1B, [state]">+25% y/y [Q3 FY2026], a Q4 guide of $10.25B (+51% y/y) and [guidance]">$4.02 EPS [guidance], plans to double manufacturing capacity by 2028 [web], non-GAAP gross margin of approximately 50.4% in Q4 [web], opex of about $1.58B [web], and Semiconductor Systems growth above the prior >[state]">30% CY26 forecast [web]. Stock reaction was [earnings recap agent]">-5.1% on the day as the market weighed supply constraints [earnings recap agent].
1. Customers are pre-paying — and AMAT is carrying the paper (changes the read: supports the guide, dents near-term cash). The 10-Q says DSO rose to 77 days because of down-payment invoices not yet recognized as revenue and reduced receivables factoring [SEC filings]. Receivables jumped to [Q3 FY2026]">$7.7B from [Q2 FY2026]">$6.4B in Q2 and [Q3 FY2025]">$5.8B a year ago [Q3 FY2026, Q2 FY2026, Q3 FY2025], and nine-month operating cash flow of $5.6B was "partially offset by a higher accounts receivable balance and higher vendor payments" [SEC filings]. Contract liabilities (customer deposits — cash or invoices for tools not yet delivered) stand at $3.3B, up [10-Q notes]">$698M year-to-date, versus a $379M decline over the same period last year [10-Q notes]. Read-through: Q4's record guide has committed customer money behind it — a demand-shortfall risk is low. But that $3.3B is ~1.0 month of guided Q4 revenue and ~[watch]">1.1 months of Q3 revenue, so the forward book is growing with sales, not ahead of them — the "forward book re-accelerating" trigger [watch] is not tripped yet. And because deposits are invoiced, not all collected, and factoring was cut, Q4 revenue will convert to cash more slowly than the P&L suggests.
2. Inventory is getting leaner relative to sales — supply, not demand, is the constraint. Inventories rose to [Q2 FY2024]">$6.6B from [Q3 FY2025]">$5.8B a year ago [Q3 FY2026, Q3 FY2025], but relative to quarterly revenue they fell from ~0.79x to ~0.73x. Customer-service spares are [10-Q notes]">$1.9B vs [Q3 FY2025]">$1.8B [10-Q notes]. The call framed constraints as customers' cleanroom space [web]; the balance sheet shows AMAT's own buffer is thinning while it commits to ~$10.25B of shipments. That is the main way the guide can be missed — execution, not orders — and it's why the ±[Q2 FY2024]">$500M band matters.
3. Capital allocation quietly shifted (changes the read of the EPS guide). [Q1 FY2025]">$1.2B of 3.300% notes due 2027 moved to current; non-current debt fell to [10-Q notes]">$5.2B from [Q2 FY2026]">$6.5B [10-Q notes]. Buybacks slowed to [Q1 FY2026]">$400M in Q3 and [Q3 FY2026]">$1.1B YTD, versus [Q3 FY2025]">$4.0B in the prior-year nine months, with [10-Q notes]">$12.8B authorization unused [10-Q notes]; insiders were net sellers of 285,588 shares over six months with zero open-market buys [state]. Cash and investments total $14.5B [10-Q notes], so the 2027 maturity is a non-event for solvency — but the [guidance]">$4.02 EPS guide is being carried by operating leverage, not share-count shrink, and cash is going to working capital and the capacity build rather than to shareholders. Flat buybacks plus [Q1 FY2025]">$1.2B of refinancing in FY2027 is a modest EPS headwind the call didn't dwell on.
4. Margin drivers are named more explicitly. The 10-Q credits "increases in average selling prices and lower material and manufacturing costs" alongside mix [SEC filings]. That corroborates the call's value-based-pricing framing [web] and the [state]">50.3% Q3 gross margin [state], comfortably above the [watch]">46.9% watch threshold [watch]. Q4's 50.4% guide [web] being roughly flat despite higher volume reflects lower-margin Display mix [web?] — single-source explanation, treat as a lead.
Items that don't change the guide read but frame it. The [10-Q notes]">$253M BIS export-controls settlement was paid in Q2 and hit the Semiconductor Systems segment [10-Q notes]; it carries audit obligations through 2028 and a suspended three-year denial order [web] — a tail risk to watch, not a Q4 factor, and non-GAAP EPS excludes it. Non-US revenue fell to [10-Q notes]">85% from 88%, and Asia Pacific to [10-Q notes]">80% of Q3 revenue from [10-Q notes]">89% a year ago [SEC filings, 10-Q notes] — a US-fab mix shift the call didn't quantify. The Q3 GAAP tax rate of [SEC filings]">12.7% ([SEC filings]">12.9% YTD) was lifted by a valuation allowance on CAMT credits [SEC filings]; GAAP net income slipped to [Q3 FY2026]">$2.5B from [Q2 FY2026]">$2.8B in Q2 despite [Q1 FY2025]">$1.2B more revenue [Q3 FY2026, Q2 FY2026], with [Q1 FY2025]">$1.2B of unrealized investment gains sitting in the portfolio [10-Q notes] — reason to anchor on non-GAAP EPS, which the guide uses. Restructuring reserve is down to [10-Q notes]">$12M from [10-Q notes]">$165M [10-Q notes], so the Q1 opex miss ([SEC-derived facts]">$1.60B vs $1.33B guided) [SEC-derived facts] was the cost-action tail, and the $1.58B Q4 opex guide [web] is the clean run-rate.
What to watch: Q4 receivables and contract liabilities versus revenue (does the deposit book grow faster than sales?), inventory/revenue below ~0.7x, buyback pace against the [10-Q notes]">$12.8B authorization, and Semiconductor Systems delivering the ~$7.9B guided [earnings recap agent].
Also disclosed, not material to this question: [10-Q notes]">$293M parent bank guarantees / [10-Q n">$282M maximum guarantee exposure; [10-Q notes]">$57M restricted cash; immaterial private-equity impairments; new FASB standards on income-tax disclosure, government grants and environmental credits.
Bottom line: The 10-Q says demand isn't the risk to the Q4 guide — pre-paid orders back it — but cash conversion, supply execution and a paused buyback are, so treat $10.25B as credible and [guidance]">$4.02 as slightly more dependent on operations than the call implied.
Sources (most reliable first): Applied Mat Q3 2026 Earnings Call Transcript - Applied Materials (NASDAQ:AMAT) - Benzinga [src:aggregator] · Applied Materials (AMAT) Q3 2026 Earnings Call Transcript [src:unverified]
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) mostly confirms the call, but four things in the notes either weren't said on the call or were framed differently – and two of them genuinely change how to read the $10.25B ± $500M / $4.02 ± $0.20 Q4 guide .
1. Customers are pre-paying – and AMAT is carrying the paper (changes the read: supports the guide, dents near-term cash). The 10-Q says DSO rose to 77 days because of down-payment invoices not yet recognized as revenue and reduced receivables factoring . Receivables jumped to $7.7B from $6.4B in Q2 and $5.8B a year ago , and nine-month operating cash flow of $5.6B was "partially offset by a higher accounts receivable balance and higher vendor payments" . Contract liabilities (customer deposits – cash or invoices for tools not yet delivered) stand at $3.3B, up $698M year-to-date, versus a $379M decline over the same period last year . Read-through: Q4's record guide has committed customer money behind it – a demand-shortfall risk is low. But that $3.3B is ~1.0 month of guided Q4 revenue and ~1.1 months of Q3 revenue, so the forward book is growing with sales, not ahead of them – the "forward book re-accelerating" trigger is not tripped yet. And because deposits are invoiced, not all collected, and factoring was cut, Q4 revenue will convert to cash more slowly than the P&L suggests.
1. Customers are pre-paying – and AMAT is carrying the paper (changes the read: supports the guide, dents near-term cash). The 10-Q says DSO rose to 77 days because of down-payment invoices not yet recognized as revenue and reduced receivables factoring . Receivables jumped to $7.7B from $6.4B in Q2 and $5.8B a year ago , and nine-month operating cash flow of $5.6B was "partially offset by a higher accounts receivable balance and higher vendor payments" . Contract liabilities (customer deposits – cash or invoices for tools not yet delivered) stand at $3.3B, up $698M year-to-date, versus a $379M decline over the same period last year . Read-through: Q4's record guide has committed customer money behind it – a demand-shortfall risk is low. But that $3.3B is ~1.0 month of guided Q4 revenue and ~1.1 months of Q3 revenue, so the forward book is growing with sales, not ahead of them – the "forward book re-accelerating" trigger is not tripped yet. And because deposits are invoiced, not all collected, and factoring was cut, Q4 revenue will convert to cash more slowly than the P&L suggests.
3. Capital allocation quietly shifted (changes the read of the EPS guide). $1.2B of 3.300% notes due 2027 moved to current; non-current debt fell to $5.2B from $6.5B . Buybacks slowed to $400M in Q3 and $1.1B YTD, versus $4.0B in the prior-year nine months, with $12.8B authorization unused ; insiders were net sellers of 285,588 shares over six months with zero open-market buys . Cash and investments total $14.5B , so the 2027 maturity is a non-event for solvency – but the $4.02 EPS guide is being carried by operating leverage, not share-count shrink, and cash is going to working capital and the capacity build rather than to shareholders. Flat buybacks plus $1.2B of refinancing in FY2027 is a modest EPS headwind the call didn't dwell on.
3. Capital allocation quietly shifted (changes the read of the EPS guide). $1.2B of 3.300% notes due 2027 moved to current; non-current debt fell to $5.2B from $6.5B . Buybacks slowed to $400M in Q3 and $1.1B YTD, versus $4.0B in the prior-year nine months, with $12.8B authorization unused ; insiders were net sellers of 285,588 shares over six months with zero open-market buys . Cash and investments total $14.5B , so the 2027 maturity is a non-event for solvency – but the $4.02 EPS guide is being carried by operating leverage, not share-count shrink, and cash is going to working capital and the capacity build rather than to shareholders. Flat buybacks plus $1.2B of refinancing in FY2027 is a modest EPS headwind the call didn't dwell on.
Items that don't change the guide read but frame it. The $253M BIS export-controls settlement was paid in Q2 and hit the Semiconductor Systems segment ; it carries audit obligations through 2028 and a suspended three-year denial order – a tail risk to watch, not a Q4 factor, and non-GAAP EPS excludes it. Non-US revenue fell to 85% from 88%, and Asia Pacific to 80% of Q3 revenue from 89% a year ago – a US-fab mix shift the call didn't quantify. The Q3 GAAP tax rate of 12.7% (12.9% YTD) was lifted by a valuation allowance on CAMT credits ; GAAP net income slipped to $2.5B from $2.8B in Q2 despite $1.2B more revenue , with $1.2B of unrealized investment gains sitting in the portfolio – reason to anchor on non-GAAP EPS, which the guide uses. Restructuring reserve is down to $12M from $165M , so the Q1 opex miss ($1.60B vs $1.33B guided) was the cost-action tail, and the $1.58B Q4 opex guide is the clean run-rate.
4. Margin drivers are named more explicitly. The 10-Q credits "increases in average selling prices and lower material and manufacturing costs" alongside mix . That corroborates the call's value-based-pricing framing and the 50.3% Q3 gross margin , comfortably above the 46.9% watch threshold . Q4's 50.4% guide being roughly flat despite higher volume reflects lower-margin Display mix – single-source explanation, treat as a lead.
Items that don't change the guide read but frame it. The $253M BIS export-controls settlement was paid in Q2 and hit the Semiconductor Systems segment ; it carries audit obligations through 2028 and a suspended three-year denial order – a tail risk to watch, not a Q4 factor, and non-GAAP EPS excludes it. Non-US revenue fell to 85% from 88%, and Asia Pacific to 80% of Q3 revenue from 89% a year ago – a US-fab mix shift the call didn't quantify. The Q3 GAAP tax rate of 12.7% (12.9% YTD) was lifted by a valuation allowance on CAMT credits ; GAAP net income slipped to $2.5B from $2.8B in Q2 despite $1.2B more revenue , with $1.2B of unrealized investment gains sitting in the portfolio – reason to anchor on non-GAAP EPS, which the guide uses. Restructuring reserve is down to $12M from $165M , so the Q1 opex miss ($1.60B vs $1.33B guided) was the cost-action tail, and the $1.58B Q4 opex guide is the clean run-rate.
Short answer: the 10-Q (filed 2026-08-20, a week after the 2026-08-13 call, not the same day) mostly confirms the call, but four things in the notes either weren't said on the call or were framed differently – and two of them genuinely change how to read the $10.25B ± $500M / $4.02 ± $0.20 Q4 guide .
QuestionStress test: if Q3 FY2027 gross margin prints at the low end of the guided range and 30% of the AI-cloud capacity commitments end up unused, what happens to FY2028 EPS, free cash flow and the valuation, and which of the risks on watch would that trip?
Rescored under proof-v2. The earlier overall ranking is withdrawn because its numeric and coverage checks were unreliable. Inspect the checklist passages and complete answers below; content coverage and recorded cost do not establish factual accuracy.
I'll pull the latest guidance, commitments and consensus numbers first.Headline: the two shocks as literally specified cost ~$0.15–0.20 of FY2028 non-GAAP EPS (~[state]">1%) and <[guidance]">2% of FCF — nearly a rounding error. The damage is in what they signal. If the low-end margin means memory inflation isn't being passed through, and unused cloud capacity means neoclouds can't sell what they've bought, the read-through is more like $1.00–1.50 of FY2028 EPS and a de-rating to ~$180–195/share vs. ~[state]">$230 today.
FY28 revenue ~$680B, GM 72.5%, non-GAAP opex ~$46B, tax 17%, ~[Q2 FY2026]">23.8B shares → non-GAAP EPS ≈ $15.50, net income ≈ $370B. FCF conversion has been running well below net income on Vera Rubin working-capital build (60% of FCF returned YTD vs. ~$46B returned in H1 implies H1 FCF of only ~$77B against ~$99B non-GAAP net income). Apply ~80% conversion → FY28 FCF ≈ $295B. At [state]">$230 that's ~14.8x FY28 EPS, ~19x FCF.
| Base | Mild (literal shocks) | Severe (read-through) | |
|---|---|---|---|
| FY28 non-GAAP EPS | ~$15.50 | ~$15.30 (-1%) | ~$14.00–14.20 (-9%) |
| FY28 FCF | ~$295B | ~$290B | ~$265–270B (working-capital relief offsets part) |
| Multiple | 14.8x | 14.5x | 13x (narrative breaks from "supply-constrained" to "demand-checked") |
| Implied price | ~[state]">$230 | ~$220 | ~$180–195 |
Note in the severe case the multiple does more damage than the EPS. A 13x forward on a company still growing [state]">50%+ is not punitive — it's what the market pays when the incremental buyer of GPUs is being financed by the seller. The 52-week low is $164.07; the severe case doesn't quite get there but removes the entire post-print rally.
Bottom line: Taken literally, the stress costs ~[state]">1% of FY28 EPS and is a non-event; taken as the signal it would actually be — memory inflation not being passed through and NVDA backstopping demand its customers can't sell — it's a ~$1.00–1.50 EPS cut and a de-rate to ~$180–195, with the circular-financing and ACIE-credit risks moving from theoretical to live.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
- Q3 FY27: revenue $108.0 billion ±2%, non-GAAP gross margin 74% ±50 basis points; opex approximately $9.2 billion GAAP and $9.0 billion non-GAAP. Low end = 73.5%. - Margin path: "We expect margins to bottom in Q4 in the 71–72% range before settling at 72–73% in fiscal year 2028 as executed price increases take effect in Q1." Driver: memory scarcity, with prices expected to rise further into next year. - FY28 revenue: a "preliminary expectation" of ~70% growth, described as "a supply-constrained outlook" – implying around $690 billion vs. a Street consensus of $573.5 billion at the time. - Cloud commitments (10-Q, July 26, 2026): cloud service agreements of $29B total – $3B remainder of FY27, $8B FY28, $7B FY29, $6B FY30, $4B FY31, $1B thereafter. The new wrinkle: NVIDIA now enters agreements where it commits to purchase cloud capacity if AI cloud partners cannot sell it to third parties, exposing the company to potential underutilization risk. - Stock: ~$229.49, market cap ~$5.67T, P/E 29.12. Q2 non-GAAP EPS was $2.22 on net income of $53.95B → ~24.3B diluted shares.
- Direct: 50bp × $108B = $540M gross profit, ~$450M after tax, ~$0.02 EPS. Irrelevant. - Read-through: a low-end print in the first quarter of the step-down means the memory curve is steeper than the CFO's Q4 71–72% trough assumed. Reasonable mapping: Q4 trough → ~70.5%, FY28 "settle" → 71.5–72% instead of 72.5%, because the price increases are fixed while input costs aren't. 50–100bp on $680B = $3.4–6.8B gross profit → $0.12–0.24 EPS, ~$3–6B FCF.
1. Memory cost/margin trajectory – already the flagged negative; a low-end Q3 confirms the 72–73% FY28 floor is at risk, and that "executed price increases" may not be enough. 2. Circular financing / demand backstop – the cloud-capacity commitments are now explicitly a put option NVDA has written to its customers; add guarantees capped at $105B for data center leases at SB Energy's campus for OpenAI, 20-year terms, and $25B of equity investment commitments. Idle capacity is the first observable sign this structure is propping demand. 3. Neocloud/ACIE credit and concentration – three direct customers were 30%, 18%, and 16% of accounts receivable; ACIE customers are the weakest balance sheets in the chain. Watch DSOs. 4. Supply commitments outrunning demand – supply and capacity commitments surged to $279B as of July 26, 2026, up from $119B last quarter. If ACIE softens, that flips from moat to inventory-provision risk (already $0.8B and $2.3B of inventory provisions in Q1 FY27 and Q1 FY26). 5. Hyperscaler cash strain – Amazon and Alphabet both reported negative free cash flow for the second quarter; unused neocloud capacity plus stretched hyperscalers means both demand legs wobble at once. 6. Pricing power – the FY28 margin recovery is entirely price-driven. You cannot push through GPU price increases into a market where your customers' racks are 30% empty.
1. Memory cost/margin trajectory – already the flagged negative; a low-end Q3 confirms the 72–73% FY28 floor is at risk, and that "executed price increases" may not be enough. 2. Circular financing / demand backstop – the cloud-capacity commitments are now explicitly a put option NVDA has written to its customers; add guarantees capped at $105B for data center leases at SB Energy's campus for OpenAI, 20-year terms, and $25B of equity investment commitments. Idle capacity is the first observable sign this structure is propping demand. 3. Neocloud/ACIE credit and concentration – three direct customers were 30%, 18%, and 16% of accounts receivable; ACIE customers are the weakest balance sheets in the chain. Watch DSOs. 4. Supply commitments outrunning demand – supply and capacity commitments surged to $279B as of July 26, 2026, up from $119B last quarter. If ACIE softens, that flips from moat to inventory-provision risk (already $0.8B and $2.3B of inventory provisions in Q1 FY27 and Q1 FY26). 5. Hyperscaler cash strain – Amazon and Alphabet both reported negative free cash flow for the second quarter; unused neocloud capacity plus stretched hyperscalers means both demand legs wobble at once. 6. Pricing power – the FY28 margin recovery is entirely price-driven. You cannot push through GPU price increases into a market where your customers' racks are 30% empty.
Short answer: the arithmetic hit is small — roughly 2–3% off FY2028 EPS and a low-single-digit-billion dent to free cash flow — but the signal would be large, because both events point at the same thing: demand not matching the commitments NVIDIA has stacked up. That is what the valuation, not the EPS line, is exposed to.
1. The gross-margin leg (FY2027 event, FY2028 read-through)
Q3 FY2027 (quarter ending late October 2026) is guided to a GAAP and non-GAAP gross margin of [guidance]">74.0% ±50bps on revenue of [guidance]">$108.0B ±[guidance]">2% (guided, not reported) [guidance]. Printing the low end, 73.5%, costs about $0.54B of gross profit (50bps × [guidance]">$108B), or roughly $0.45B after the guided 16–18% tax rate — about $0.02 a share on ~24.3B shares (derived: [state]">$5,598.7B market cap ÷ [state]">$230.36 price) [state]. That is a FY2027 rounding error.
The FY2028 damage is the read-through. Management already reset margins: Q4 FY2027 (quarter ending late January 2027) guided to bottom at 71–72%, FY2028 at 72–73%, because memory price increases "exceeded prior expectations and are headed even higher" [earnings recap agent]. A low-end Q3 says the memory curve is still running ahead of them, so the FY2028 range probably slips too. Sensitivity on consensus FY2028 revenue of $685.85B (analyst estimate) [estimate]: each 100bps of gross margin is ~$6.9B of gross profit, ~$5.7B after tax, ~$0.23/share — about 1.5% of consensus FY2028 EPS of $15.64 [estimate]. A 150bps miss versus the 72.5% midpoint would be ~$0.35/share, ~[guidance]">2%.
Context that matters: the reported Q2 FY2027 (quarter ended July 2026) GAAP gross margin was [state]">75.0%, up from [Q1 FY2027]">74.9% in Q1 FY2027 and [Q2 FY2026]">72.4% in Q2 FY2026 [Q2 FY2027]. So the guided path is a ~350–400bps fall from a peak, already partly priced — shares rose [earnings recap agent]">8.7% on the print despite the reset [earnings recap agent].
2. The unused-capacity leg
The 10-Q lists "AI cloud agreements" commitments of [10-Q notes]">$36.0B: $6.0B in year 1, [10-Q notes]">$8.0B year 2, [10-Q notes]">$7.0B year 3, $6.0B year 4, [guidance]">$9.0B after (reported) [10-Q notes]. These are NVIDIA's obligations to buy capacity from AI clouds, so unused capacity is a stranded cost to NVIDIA, not lost revenue. [state]">30% unused = ~$1.8B in FY2028 and ~$10.8B across the schedule. (If you also count the separate [10-Q notes]">$29.0B of cloud service agreements, year-1 [10-Q notes]">$8.0B, add ~$2.4B — call it up to ~$4.2B in FY2028 [10-Q notes].) After tax, that is another ~$0.06–$0.14/share, under [state]">1% of EPS. For FCF the cash still leaves — but against Q2 FY2027 reported net income of $59.7B and [Q2 FY2027]">$39.0B of year-to-date buybacks, $2–4B is manageable [Q2 FY2027]. There is no FY2028 FCF consensus figure here to net it against directly.
Combined arithmetic: roughly $0.30–$0.50 off $15.64, i.e. ~2–3%; forward P/E drifts from [state]">24.9x to ~25.5x [state]. Nobody re-rates a stock on that.
3. Why the valuation reacts anyway
The price embeds ~[state]">29.9% annual free-cash-flow growth for a decade (modeled, as of 2026-09-08; the older valuation run at $220.78 showed 29.4%) [state]. Unused neocloud capacity is the first hard evidence that demand can undershoot commitments — and NVIDIA's commitments are now enormous: supply and capacity commitments of [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279.0B, up from [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119.0B a quarter earlier, of which [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$92.0B falls in the rest of this fiscal year and [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$87.0B in year 1 (reported) [10-Q notes]; $105.0B of guarantees on SB Energy land, power and shell buildouts entered in August 2026 [10-Q notes]; and management's own acknowledgment of nearly $50B of balance-sheet exposure to frontier labs without investment-grade credit [earnings recap agent]. Balance-sheet tells already lean the same way: receivables jumped to [Q2 FY2027]">$63.1B (DSO ~60 days) from [Q1 FY2027]">$40.7B (~45 days) in Q1 FY2027, inventories to [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B, and total debt to [Q3 FY2026]">$33.4B from [Q1 FY2026]">$8.5B after [Q2 FY2027]">$24.9B of issuance [Q2 FY2027]. Five customers are 22%, 14%, 13%, [state]">11% and [watch]">10% of receivables [10-Q notes]. In that setting, a [state]">30% utilisation shortfall makes the market ask whether the [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B of supply is take-or-pay against demand that isn't there — and whether "[earnings recap agent]">70% FY2028 growth, supply-constrained, true demand 100%" [earnings recap agent] still holds. The relative-value floor is already thin: peer-median P/S implies ~[state]">-69.5% downside, a trigger that has FIRED [state].
4. Which watches trip
Also disclosed, not material to this question: [10-Q notes]">$25.0B data-center leases not yet commenced, [10-Q notes]">$25.0B equity investment commitments, 38% of Q2 revenue from customers headquartered outside the US, Hugging Face acquisition integration risk [10-Q notes] [risk].
What to watch: Q3 FY2027 print (late November 2026) gross margin versus 73.5% and any change to the 71–72% Q4 floor; CoreWeave CDS crossing [watch]">600bps; the supply-commitment line in the Q3 10-Q — does [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B keep climbing while receivables DSO stays ~60 days.
Bottom line: This stress case shaves ~2–3% off FY2028 EPS and a few billion of FCF — trivial against a $59.7B quarter — but it would convert the "supply-constrained" story into a "commitment-heavy, demand-uncertain" one, and with the stock priced for ~[state]">30% FCF growth for ten years, that narrative shift, not the arithmetic, is where the downside sits.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Q3 FY2027 (quarter ending late October 2026) is guided to a GAAP and non-GAAP gross margin of 74% ±50bps on revenue of $108.0B ±2% (guided, not reported) . Printing the low end, 73.5%, costs about $0.54B of gross profit (50bps × $108B), or roughly $0.45B after the guided 16–18% tax rate – about $0.02 a share on ~24.3B shares (derived: $5,598.7B market cap ÷ $230.36 price) . That is a FY2027 rounding error.
Q3 FY2027 (quarter ending late October 2026) is guided to a GAAP and non-GAAP gross margin of 74% ±50bps on revenue of $108.0B ±2% (guided, not reported) . Printing the low end, 73.5%, costs about $0.54B of gross profit (50bps × $108B), or roughly $0.45B after the guided 16–18% tax rate – about $0.02 a share on ~24.3B shares (derived: $5,598.7B market cap ÷ $230.36 price) . That is a FY2027 rounding error.
The 10-Q lists "AI cloud agreements" commitments of $36.0B: $6.0B in year 1, $8.0B year 2, $7.0B year 3, $6.0B year 4, $9.0B after (reported) . These are NVIDIA's obligations to buy capacity from AI clouds, so unused capacity is a stranded cost to NVIDIA, not lost revenue. 30% unused = ~$1.8B in FY2028 and ~$10.8B across the schedule. (If you also count the separate $29.0B of cloud service agreements, year-1 $8.0B, add ~$2.4B – call it up to ~$4.2B in FY2028 .) After tax, that is another ~$0.06–$0.14/share, under 1% of EPS. For FCF the cash still leaves – but against Q2 FY2027 reported net income of $59.7B and $39.0B of year-to-date buybacks, $2–4B is manageable Q2 FY2027. There is no FY2028 FCF consensus figure here to net it against directly.
The FY2028 damage is the read-through. Management already reset margins: Q4 FY2027 (quarter ending late January 2027) guided to bottom at 71–72%, FY2028 at 72–73%, because memory price increases "exceeded prior expectations and are headed even higher" . A low-end Q3 says the memory curve is still running ahead of them, so the FY2028 range probably slips too. Sensitivity on consensus FY2028 revenue of $685.85B (analyst estimate) : each 100bps of gross margin is ~$6.9B of gross profit, ~$5.7B after tax, ~$0.23/share – about 1.5% of consensus FY2028 EPS of $15.64 . A 150bps miss versus the 72.5% midpoint would be ~$0.35/share, ~2%.
The FY2028 damage is the read-through. Management already reset margins: Q4 FY2027 (quarter ending late January 2027) guided to bottom at 71–72%, FY2028 at 72–73%, because memory price increases "exceeded prior expectations and are headed even higher" . A low-end Q3 says the memory curve is still running ahead of them, so the FY2028 range probably slips too. Sensitivity on consensus FY2028 revenue of $685.85B (analyst estimate) : each 100bps of gross margin is ~$6.9B of gross profit, ~$5.7B after tax, ~$0.23/share – about 1.5% of consensus FY2028 EPS of $15.64 . A 150bps miss versus the 72.5% midpoint would be ~$0.35/share, ~2%.
- gmslip (<73%): a 73.5% print does not trip it, but Q4's 71–72% guide is already scheduled to fire; a low-end Q3 raises the odds Q4 lands below 71% . - memorypricing: escalates from WATCH to confirmed – the driver of gmslip . - coreweavecds: already FIRED once (881bps peak, ~452bps now); unused neocloud capacity is exactly what would push it back above 600bps . - openaicircularfinancing: unused vendor-financed capacity is the escalation case for the ~$250B deal . - hyperscalercapex2027: not tripped by neoclouds alone, but this is the one to check next – a coordinated 2027 cut "flips the thesis" . - reversedcfbar (>32%): if FCF forecasts fall while price holds, implied growth rises toward the bar . - Risk factors engaged: Supply chain ("demand forecasting errors cause supply-demand mismatches"), Customer concentration & demand, and Liquidity/debt ("extended customer payment terms") .
The 10-Q lists "AI cloud agreements" commitments of $36.0B: $6.0B in year 1, $8.0B year 2, $7.0B year 3, $6.0B year 4, $9.0B after (reported) . These are NVIDIA's obligations to buy capacity from AI clouds, so unused capacity is a stranded cost to NVIDIA, not lost revenue. 30% unused = ~$1.8B in FY2028 and ~$10.8B across the schedule. (If you also count the separate $29.0B of cloud service agreements, year-1 $8.0B, add ~$2.4B – call it up to ~$4.2B in FY2028 .) After tax, that is another ~$0.06–$0.14/share, under 1% of EPS. For FCF the cash still leaves – but against Q2 FY2027 reported net income of $59.7B and $39.0B of year-to-date buybacks, $2–4B is manageable Q2 FY2027. There is no FY2028 FCF consensus figure here to net it against directly.
- gmslip (<73%): a 73.5% print does not trip it, but Q4's 71–72% guide is already scheduled to fire; a low-end Q3 raises the odds Q4 lands below 71% . - memorypricing: escalates from WATCH to confirmed – the driver of gmslip . - coreweavecds: already FIRED once (881bps peak, ~452bps now); unused neocloud capacity is exactly what would push it back above 600bps . - openaicircularfinancing: unused vendor-financed capacity is the escalation case for the ~$250B deal . - hyperscalercapex2027: not tripped by neoclouds alone, but this is the one to check next – a coordinated 2027 cut "flips the thesis" . - reversedcfbar (>32%): if FCF forecasts fall while price holds, implied growth rises toward the bar . - Risk factors engaged: Supply chain ("demand forecasting errors cause supply-demand mismatches"), Customer concentration & demand, and Liquidity/debt ("extended customer payment terms") .
- gmslip (<73%): a 73.5% print does not trip it, but Q4's 71–72% guide is already scheduled to fire; a low-end Q3 raises the odds Q4 lands below 71% . - memorypricing: escalates from WATCH to confirmed – the driver of gmslip . - coreweavecds: already FIRED once (881bps peak, ~452bps now); unused neocloud capacity is exactly what would push it back above 600bps . - openaicircularfinancing: unused vendor-financed capacity is the escalation case for the ~$250B deal . - hyperscalercapex2027: not tripped by neoclouds alone, but this is the one to check next – a coordinated 2027 cut "flips the thesis" . - reversedcfbar (>32%): if FCF forecasts fall while price holds, implied growth rises toward the bar . - Risk factors engaged: Supply chain ("demand forecasting errors cause supply-demand mismatches"), Customer concentration & demand, and Liquidity/debt ("extended customer payment terms") .
- gmslip (<73%): a 73.5% print does not trip it, but Q4's 71–72% guide is already scheduled to fire; a low-end Q3 raises the odds Q4 lands below 71% . - memorypricing: escalates from WATCH to confirmed – the driver of gmslip . - coreweavecds: already FIRED once (881bps peak, ~452bps now); unused neocloud capacity is exactly what would push it back above 600bps . - openaicircularfinancing: unused vendor-financed capacity is the escalation case for the ~$250B deal . - hyperscalercapex2027: not tripped by neoclouds alone, but this is the one to check next – a coordinated 2027 cut "flips the thesis" . - reversedcfbar (>32%): if FCF forecasts fall while price holds, implied growth rises toward the bar . - Risk factors engaged: Supply chain ("demand forecasting errors cause supply-demand mismatches"), Customer concentration & demand, and Liquidity/debt ("extended customer payment terms") .
The price embeds ~29.9% annual free-cash-flow growth for a decade (modeled, as of 2026-09-08; the older valuation run at $220.78 showed 29.4%) . Unused neocloud capacity is the first hard evidence that demand can undershoot commitments – and NVIDIA's commitments are now enormous: supply and capacity commitments of $279.0B, up from $119.0B a quarter earlier, of which $92.0B falls in the rest of this fiscal year and $87.0B in year 1 (reported) ; $105.0B of guarantees on SB Energy land, power and shell buildouts entered in August 2026 ; and management's own acknowledgment of nearly $50B of balance-sheet exposure to frontier labs without investment-grade credit . Balance-sheet tells already lean the same way: receivables jumped to $63.1B (DSO ~60 days) from $40.7B (~45 days) in Q1 FY2027, inventories to $31.6B from $25.8B, and total debt to $33.4B from $8.5B after $24.9B of issuance Q2 FY2027. Five customers are 22%, 14%, 13%, 11% and 10% of receivables . In that setting, a 30% utilisation shortfall makes the market ask whether the $279B of supply is take-or-pay against demand that isn't there – and whether "70% FY2028 growth, supply-constrained, true demand 100%" still holds. The relative-value floor is already thin: peer-median P/S implies ~-69.5% downside, a trigger that has FIRED .
Short answer: the direct P&L damage from that stress is small — roughly 1–4% off FY2028 EPS — but the signal it sends would do far more to the multiple than the charges do to earnings, and it would trip three watch items that are already primed.
1. Sizing the two shocks against FY2028 consensus
Baseline: FY2028 consensus EPS is $15.64 on revenue of ~$685.9B [estimate], both revised up last week [estimate]. Market cap of [state]">$5,598.7B at [state]">$230.36 [state] implies ~24.3B shares, so consensus embeds roughly $380B of net income — a ~55% net margin, in line with FY2026's [state]">55.6% [state].
Shock A — Q3 FY2027 gross margin at the low end (73.5%). Guidance is [guidance]">74.0% ±50bps on [guidance]">$108.0B revenue [guidance]. The low end costs ~[Q2 FY2026]">$0.5B of gross profit in Q3 — about $0.02/share after the 16–18% guided tax rate [guidance] — and it lands in FY2027, not FY2028. It also would not trip the gm_slip trigger, whose threshold is <73% [watch]. What matters is the read-through: management has already reset Q4 FY2027 to 71–72% and FY2028 to 72–73% because memory prices "exceeded prior expectations and are headed even higher" [earnings recap]. A low-end Q3 says the memory curve is still worse than modeled. If FY2028 lands 50–100bps below the guided band, that is ~$3.4–6.9B of gross profit on $685.9B revenue, or ~$0.12–0.23/share after tax — a 0.7–1.5% EPS haircut. Note NVDA is fabless: this hits its own COGS because HBM is a purchased input, not because it runs fabs.
Shock B — [state]">30% of AI-cloud capacity commitments unused. The 10-Q table shows AI cloud agreements of [10-Q notes]">$36.0B: [10-Q notes]">$0 in the rest of FY2027, $6.0B in FY2028, [10-Q notes]">$8B/[10-Q notes]">$7B/[Q2 FY2027]">$6B in FY2029–31 and [guidance]">$9B thereafter [10-Q notes]. The mechanism matters: these are described as agreements where NVIDIA commits to purchase cloud capacity if AI cloud partners cannot sell it to third parties, exposing the company to potential underutilization risk [web?] — i.e. a backstop, so "unused" means NVDA pays for GPUs sitting idle. On the FY2028 tranche, [state]">30% = ~$1.8B pre-tax, ~$0.06/share (-0.4% EPS). The tail case is if auditors treat the whole unused slice as an onerous commitment and force an upfront loss accrual: [state]">30% of [Q3 FY2026]">$36B = ~$10.8B pre-tax, ~$0.37/share (-2.4%). Whether that accrual would be required isn't knowable from the filing.
Combined: FY2028 EPS moves from $15.64 to roughly $15.20–15.45 in the run-rate case, ~$15.0 in the accrual case. Free cash flow takes the same ~$5–9B hit (both shocks are cash), which against a ~$380B earnings base is low-single-digit percent — but I can't give you a precise FCF figure, and the [state]">10.9% accrual ratio [state] plus DSO stretching from 45 to 60 days in one quarter [Q1 FY2027, Q2 FY2027] means cash already lags earnings before any stress.
2. Why the second-order effects dwarf the first
Thirty percent idle neocloud capacity is not a $1.8B problem; it is a statement that the "true demand is 100%, we're only constrained to [earnings recap agent]">70%" narrative [earnings recap] is wrong. That re-prices everything downstream:
3. Valuation
At [state]">$230.36 the stock is [state]">24.9x forward EPS [estimate]; on $15.2 that's ~25.7x, so a constant-multiple price effect is only ~-3%. But the reverse DCF already requires ~[state]">29.9% annual FCF growth for ten years at a 13.9% discount rate [reverse-DCF], with a [watch]">32% alarm line [watch]. If FCF estimates fall and price doesn't, implied growth rises toward that line; if the market instead stops underwriting the growth, the multiple — not the EPS — does the damage. Precedent: a mere report of the ~$250B OpenAI deal took the stock down ~4% on 2026-07-27 [watch]. The comps check ([state]">-69.5% to peer-median P/S) has already fired [state], so there is no relative-value floor nearby; the Street's [state]">49.9% target upside [estimate] would be the first thing to compress.
4. Which watch items trip
Also disclosed, not material to this question: [Q2 FY2027]">$25B equity investment commitments, [10-Q notes]">$29B cloud service agreements (mostly for NVDA's own R&D), [Q2 FY2027]">$25B data-center leases not commenced, [Q3 FY2026]">$33.4B total debt after the [Q2 FY2027]">$24.9B June issuance, 38% non-US revenue [10-Q notes, Q2 FY2027].
What to watch: the Q3 print (late Nov) — gross margin vs [guidance]">74%, DSO vs 60 days, inventory vs [Q2 FY2027]">$31.6B, and any change to the [Q3 FY2026]">$36B AI-cloud line or the FY2028 72–73% margin band.
Bottom line: Mechanically, the stress costs FY2028 EPS ~1–4% and FCF a few billion — survivable. But it would invalidate the supply-constrained-not-demand-constrained thesis, put $50B of lab exposure, $105B of guarantees and [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B of supply commitments in play, and the multiple would move long before the charges hit.
Sources (most reliable first): Nvidia 10-Q Risks & Red Flags Redline | Hudson Labs [src:unverified]
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Shock A – Q3 FY2027 gross margin at the low end (73.5%). Guidance is 74% ±50bps on $108.0B revenue . The low end costs ~$0.5B of gross profit in Q3 – about $0.02/share after the 16–18% guided tax rate – and it lands in FY2027, not FY2028. It also would not trip the gmslip trigger, whose threshold is <73% . What matters is the read-through: management has already reset Q4 FY2027 to 71–72% and FY2028 to 72–73% because memory prices "exceeded prior expectations and are headed even higher" . A low-end Q3 says the memory curve is still worse than modeled. If FY2028 lands 50–100bps below the guided band, that is ~$3.4–6.9B of gross profit on $685.9B revenue, or ~$0.12–0.23/share after tax – a 0.7–1.5% EPS haircut. Note NVDA is fabless: this hits its own COGS because HBM is a purchased input, not because it runs fabs.
Shock A – Q3 FY2027 gross margin at the low end (73.5%). Guidance is 74% ±50bps on $108.0B revenue . The low end costs ~$0.5B of gross profit in Q3 – about $0.02/share after the 16–18% guided tax rate – and it lands in FY2027, not FY2028. It also would not trip the gmslip trigger, whose threshold is <73% . What matters is the read-through: management has already reset Q4 FY2027 to 71–72% and FY2028 to 72–73% because memory prices "exceeded prior expectations and are headed even higher" . A low-end Q3 says the memory curve is still worse than modeled. If FY2028 lands 50–100bps below the guided band, that is ~$3.4–6.9B of gross profit on $685.9B revenue, or ~$0.12–0.23/share after tax – a 0.7–1.5% EPS haircut. Note NVDA is fabless: this hits its own COGS because HBM is a purchased input, not because it runs fabs.
Shock B – 30% of AI-cloud capacity commitments unused. The 10-Q table shows AI cloud agreements of $36.0B: $0 in the rest of FY2027, $6.0B in FY2028, $8B/$7B/$6B in FY2029–31 and $9B thereafter . The mechanism matters: these are described as agreements where NVIDIA commits to purchase cloud capacity if AI cloud partners cannot sell it to third parties, exposing the company to potential underutilization risk – i.e. a backstop, so "unused" means NVDA pays for GPUs sitting idle. On the FY2028 tranche, 30% = ~$1.8B pre-tax, ~$0.06/share (-0.4% EPS). The tail case is if auditors treat the whole unused slice as an onerous commitment and force an upfront loss accrual: 30% of $36B = ~$10.8B pre-tax, ~$0.37/share (-2.4%). Whether that accrual would be required isn't knowable from the filing.
Baseline: FY2028 consensus EPS is $15.64 on revenue of ~$685.9B , both revised up last week . Market cap of $5,598.7B at $230.36 implies ~24.3B shares, so consensus embeds roughly $380B of net income – a ~55% net margin, in line with FY2026's 55.6% .
Baseline: FY2028 consensus EPS is $15.64 on revenue of ~$685.9B , both revised up last week . Market cap of $5,598.7B at $230.36 implies ~24.3B shares, so consensus embeds roughly $380B of net income – a ~55% net margin, in line with FY2026's 55.6% .
Shock A – Q3 FY2027 gross margin at the low end (73.5%). Guidance is 74% ±50bps on $108.0B revenue . The low end costs ~$0.5B of gross profit in Q3 – about $0.02/share after the 16–18% guided tax rate – and it lands in FY2027, not FY2028. It also would not trip the gmslip trigger, whose threshold is <73% . What matters is the read-through: management has already reset Q4 FY2027 to 71–72% and FY2028 to 72–73% because memory prices "exceeded prior expectations and are headed even higher" . A low-end Q3 says the memory curve is still worse than modeled. If FY2028 lands 50–100bps below the guided band, that is ~$3.4–6.9B of gross profit on $685.9B revenue, or ~$0.12–0.23/share after tax – a 0.7–1.5% EPS haircut. Note NVDA is fabless: this hits its own COGS because HBM is a purchased input, not because it runs fabs.
- openaicircularfinancing (WATCH → escalation): idle backstopped capacity is vendor-financed demand proving soft . - coreweavecds (already FIRED): neocloud stress is exactly what pushes the canary back above 600bps from ~452 . - memorypricing / gmslip: Q3 at 73.5% doesn't fire gmslip, but Q4 is already scheduled to (71–72%) and a low-end Q3 raises the odds it lands below 71% . - reversedcfbar and accrualhigh move toward their thresholds via lower FCF and higher receivables/inventory ; hyperscalercapex2027 is the follow-on to watch, since neocloud weakness typically precedes hyperscaler cuts . - Risk factors engaged: Customer concentration & demand, Supply chain (forecasting), Liquidity/extended payment terms .
Thirty percent idle neocloud capacity is not a $1.8B problem; it is a statement that the "true demand is 100%, we're only constrained to 70%" narrative is wrong. That re-prices everything downstream: - The $685.9B FY2028 revenue consensus would come down – a 10% revenue cut at a 55% net margin removes ~$1.55/share, five times the mechanical charges above. - The $279B of supply commitments ($87B due in FY2028) and $31.6B of inventory, up from $25.8B a quarter ago , become the real FCF risk: NVDA has pre-bought wafers and HBM for a ramp its customers may not absorb. That is the "demand forecasting errors cause supply-demand mismatches" risk factor verbatim . - Balance-sheet exposure to the same customers: ~$50B in frontier labs without investment-grade credit ; $63.1B of receivables with five direct customers at 22/14/13/11/10% ; and the $105B SB Energy guarantees , which a single filing review says back data center leases (4.25 GW IT load) at SB Energy's PORTS Technology Campus for OpenAI . The same review puts private equity investments at $99B as of July 26, 2026 (vs. $22.3B at January 25, 2026) – unconfirmed, but if right, mark-downs there would swamp the capacity charge.
At $230.36 the stock is 24.9x forward EPS ; on $15.2 that's ~25.7x, so a constant-multiple price effect is only ~-3%. But the reverse DCF already requires ~29.9% annual FCF growth for ten years at a 13.9% discount rate , with a 32% alarm line . If FCF estimates fall and price doesn't, implied growth rises toward that line; if the market instead stops underwriting the growth, the multiple – not the EPS – does the damage. Precedent: a mere report of the ~$250B OpenAI deal took the stock down ~4% on 2026-07-27 . The comps check (-69.5% to peer-median P/S) has already fired , so there is no relative-value floor nearby; the Street's 49.9% target upside would be the first thing to compress.
Thirty percent idle neocloud capacity is not a $1.8B problem; it is a statement that the "true demand is 100%, we're only constrained to 70%" narrative is wrong. That re-prices everything downstream: - The $685.9B FY2028 revenue consensus would come down – a 10% revenue cut at a 55% net margin removes ~$1.55/share, five times the mechanical charges above. - The $279B of supply commitments ($87B due in FY2028) and $31.6B of inventory, up from $25.8B a quarter ago , become the real FCF risk: NVDA has pre-bought wafers and HBM for a ramp its customers may not absorb. That is the "demand forecasting errors cause supply-demand mismatches" risk factor verbatim . - Balance-sheet exposure to the same customers: ~$50B in frontier labs without investment-grade credit ; $63.1B of receivables with five direct customers at 22/14/13/11/10% ; and the $105B SB Energy guarantees , which a single filing review says back data center leases (4.25 GW IT load) at SB Energy's PORTS Technology Campus for OpenAI . The same review puts private equity investments at $99B as of July 26, 2026 (vs. $22.3B at January 25, 2026) – unconfirmed, but if right, mark-downs there would swamp the capacity charge.
QuestionLooking ahead to FY2028: what has to go right for NVIDIA's ~70% growth outlook to hold, what are the earliest signs it is slipping, and what should I watch over the next two quarters?
Rescored under proof-v2. The earlier overall ranking is withdrawn because its numeric and coverage checks were unreliable. Inspect the checklist passages and complete answers below; content coverage and recorded cost do not establish factual accuracy.
Recorded claims and their settling events. Clear / warning fired describes an observed risk condition, not prediction accuracy. Incomplete or unsupported tests require reader review. Hover, focus or tap a status for its meaning.
| Side | Claim | Settles at | Observed | Status |
|---|---|---|---|---|
| web search | Q3 FY2027 print: the Q4 guide needs to be at least about $118B for the FY2028 supply ramp to be on the curve.the answer's own bar; a guide below it is the answer's slip signal, so the claim is scored as a signpost, not a forecast | Q3 FY2027 release, late November 2026 | — | OpenStill pending at the last check. The settling event or required observation has not arrived yet. Any value shown is not a final result. |
| web search | Gross margin breaking below 71% would be the first crack in pricing power.clear if the recorded 71% floor holds; triggered if gross margin is lower. This is a risk condition, not forecast accuracy. | Q4 FY2027 release, late February 2027 | — | OpenStill pending at the last check. The settling event or required observation has not arrived yet. Any value shown is not a final result. |
| web search | FY2028 revenue of about $680–700B requires roughly 10–15% sequential growth every quarter for four more quarters.A four-quarter cadence claim needs four period-specific observations; the old rule checked Q1 only. | each FY2028 10-Q, first at the Q1 FY2028 print in May 2027 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| web search | The Q1 FY2028 revenue guide will be about $140B or more.The original answer called $140B+ implied, not an unconditional forecast. Review the model assumptions rather than scoring it as a prediction. | Q4 FY2027 release, late February 2027 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| web search | Hyperscaler September-quarter calls in late October give the first 2027 capex framing; benchmark against the CFO's $1.3T top-five figure.not in Finbly's stores yet; the hyperscaler_capex_2027 trigger is reviewed monthly by hand | MSFT/AMZN/META/GOOGL calls, late October 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| web search | Street consensus for FY2028 revenue is $573.5B, so the 70% guide implies a large beat.Historical consensus must be checked against the answer-time snapshot, not current estimates or an arbitrary $600B cutoff. | verifiable now | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | Consensus already models about 68% FY2028 growth ($685.9B on $408.4B), so holding the outlook is the base case, not a beat.The original growth claim needs both dated FY2027 and FY2028 consensus values; a current revenue-only cutoff is insufficient. | verifiable now; re-checked at each estimates snapshot | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | A Q3 print below 73.5% gross margin, or a Q4 guide below 71%, would say memory is eating more than modeled.The Q4 gross-margin-guide condition is not covered by B6, which tests revenue; both conditions require explicit resolution. | Q3 FY2027 release, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | Receivables outrunning revenue is the earliest slip signal: DSO stretched to ~60 days; another leg up means customers are being carried.The 65-day cutoff was editorial, not stated in the answer. Review the actual change and the accompanying claim. | Q3 FY2027 10-Q, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | Customer advances ($15.6B in the first half) are the offset; a sequential drop would mean customers stop prepaying to reserve supply.A sequential drop requires Q3 versus Q2 advances, not nine-month versus six-month cumulative totals. | Q3 FY2027 10-Q, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | CoreWeave CDS back above 600bp signals renewed neocloud stress.the metric is reviewed by hand on the board; unresolved until it carries a value | continuous; trigger board | — | OpenStill pending at the last check. The settling event or required observation has not arrived yet. Any value shown is not a final result. |
| Finbly research | The Q4 guide should still imply FY2028 on a ~70% path; a guide that does not is the slip.The $118B rule was imported from another answer; this claim needs its own explicit, reviewed test. | Q3 FY2027 release, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research | Estimate revisions are 5 up / 0 down; the first coordinated cuts to FY2028 would matter more than any headline.Coordinated downward revisions cannot be established by a current revenue level against an editorial $680B cutoff. | weekly estimates snapshots | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Q4 FY2027 is implied at roughly $122–123B from the Q3 guide and FY2027 consensus.This is an implied scenario calculation, not a forecast. The old $117–128B tolerance was not the quoted $122–123B claim. | Q3 FY2027 release, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Q3 below $108B or Q4 guided below about $120B breaks the 13–14% sequential cadence the outlook needs.The Q4-below-$120B condition is not tested by C1, whose former pass range included $117B. Review both conditions. | Q3 FY2027 release, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Another ten-day rise in DSO in Q3 with flat cash conversion is the first hard-data warning.The original warning combines DSO and cash conversion; the old rule tested only DSO. | Q3 FY2027 10-Q, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | A sequential drop in customer advances means customers are no longer prepaying to reserve supply.A sequential drop requires Q3 versus Q2 advances, not cumulative totals. | Q3 FY2027 10-Q, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Samsung reportedly reached about 80% HBM4 yield in late August (flagged unconfirmed, single source). | supplier disclosures; not in Finbly's stores | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Alphabet flagged a further 2027 capex increase; Meta and Microsoft declined to guide.resolves with the hyperscaler_capex_2027 review | hyperscaler September-quarter calls, late October 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
| Finbly research + web search | Gross margin below the 73.5% low end in Q3, or a Q4 guide below 71%, means memory is winning the pricing fight.The original warning includes a Q4 margin-guide condition that the old resolver did not test. | Q3 FY2027 release, late November 2026 | — | Reader checkThis claim needs human review. Its evidence may be unavailable, or its recorded test may not cover the full claim. It is not automatically scored as right or wrong; see the note beside the claim. |
The number: ~[earnings recap agent]">70% FY28 growth ≈ $680–700B revenue, and it's a supply ceiling, not a demand forecast.
CFO Colette Kress said on a call with analysts that Nvidia expects fiscal 2028 revenue growth of [earnings recap agent]">70%, while analysts were expecting 44%. Kress said that customer forecasts "point to our growth doubling next year," but she said guidance reflects supply constraints. Guidance for [earnings recap agent]">70% revenue growth in fiscal 2028 implies around $690 billion ahead of a Street consensus estimate of $573.5 billion. Rough run-rate math: NVDA exits FY27 around [10-Q notes]">$120B/quarter (Q3 guide [guidance]">$108B); $690B in FY28 needs an average of ~$170B/quarter, i.e. roughly 10–15% sequential growth every quarter for four more quarters. That's the bar.
What has to go right
Memory (the binding constraint). Management forecast revenue growth of roughly [earnings recap agent]">70% for fiscal 2028, noting that its production capacity was constrained by memory supply. NVDA has bought its way to the front of the line: purchase commitments to suppliers, including memory, more than doubled to [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279 billion from [Supply And Capacity Commitments] (as of 2026-04-26) = $119.0B — “Supply and capacity – We have partnered with our extensive network to secure the necessary supply and critical components needed to meet demand for the next several years, incr">$119 billion in the previous quarter. Those commitments must convert into delivered HBM4 on schedule from all three suppliers.
Packaging. HBM4 supply is constrained by advanced semiconductor packaging capacity at TSMC, and prices have risen sharply as AI infrastructure builders compete for the same supply. TSMC CoWoS confirmed at ~130,000 wpm by YE 2026 (up from ~75,000 in 2025, ~40,000 in 2024). FY28 needs another step-up beyond that.
Rubin ramp without a Blackwell-style hiccup. Vera Rubin GPU: production shipments Q3 FY27, volume ramp Q4 FY27. Vera Rubin is already in full production and ramping across CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Essentially all of FY28's incremental revenue is Rubin.
Pricing sticks, so margins trough and recover. Gross margins, while still exceptional at [state]">75% for Q2, will compress to approximately [guidance]">74% in Q3 and trough at 71% to 72% in Q4, before NVIDIA expects a recovery to 72% to 73% in fiscal 2028 as price increases in its supply agreements take effect. AI server prices will rise more than 15% on Vera Rubin and Grace Blackwell systems shipping in early 2027. Customers accepting +15% ASPs is what turns [earnings recap agent]">70% unit-constrained growth into [earnings recap agent]">70% revenue growth.
Customer capex and financing hold. Kress projected $1.3 trillion in capex from the top five hyperscalers next year and a cloud backlog of over $2 trillion. The non-hyperscale half is increasingly financed: with its customers' spending outpacing their cash generation, Nvidia is helping secure over [debate agent]">$500 billion in capital from third parties. The company guarantees minimum revenue via take-or-pay contracts for part of the capacity, then earns from hardware sales and again from the rental upside. That works while capital is available; with Nvidia's financing model dependent on access to capital, rising long-term interest rates could tighten access for the labs and the neoclouds.
Note China is not in the number — NVIDIA is not assuming any Data Center compute revenue from China in its outlook. That's upside optionality, not a dependency.
Earliest signs it's slipping
Calendar for the next two quarters
Bottom line: The [earnings recap agent]">70% is a memory-and-packaging capacity number with demand running ~100%+, so FY28 breaks on supply execution and pricing stickiness first — watch the Q4 guide (≥$118B), GM holding the 71–72% floor, and hyperscaler 2027 capex framing in late October; demand erosion via custom silicon or neocloud credit stress is the slower, second-order risk.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
The number: ~70% FY28 growth ≈ $680–700B revenue, and it's a supply ceiling, not a demand forecast.
1. Memory (the binding constraint). Management forecast revenue growth of roughly 70% for fiscal 2028, noting that its production capacity was constrained by memory supply. NVDA has bought its way to the front of the line: purchase commitments to suppliers, including memory, more than doubled to $279 billion from $119 billion in the previous quarter. Those commitments must convert into delivered HBM4 on schedule from all three suppliers.
1. Memory (the binding constraint). Management forecast revenue growth of roughly 70% for fiscal 2028, noting that its production capacity was constrained by memory supply. NVDA has bought its way to the front of the line: purchase commitments to suppliers, including memory, more than doubled to $279 billion from $119 billion in the previous quarter. Those commitments must convert into delivered HBM4 on schedule from all three suppliers.
3. Rubin ramp without a Blackwell-style hiccup. Vera Rubin GPU: production shipments Q3 FY27, volume ramp Q4 FY27. Vera Rubin is already in full production and ramping across CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Essentially all of FY28's incremental revenue is Rubin.
- Late Sept: Micron FQ4 – HBM4 sold-out status and 2027 contract pricing; a rollover in memory prices would actually be good for NVDA margins but might signal cooling demand. - Mid-Oct: TSMC Q3 – CoWoS capacity adds for 2027; any capex cut is a red flag for FY28 supply. - Late Oct / early Nov: MSFT, GOOGL, AMZN, META Q3 – first 2027 capex framing. Benchmark it against Kress's $1.3T top-five figure; also watch commentary on custom accelerators vs GPUs. - ~Nov 19: NVDA Q3 FY27. Checklist: revenue vs $108B; Q4 guide (need ≥~$118B); GM ~74%; hyperscale vs ACIE split; purchase commitments; inventory; Rubin volume commentary; any change to the 70% frame. - Dec: Broadcom (custom ASIC trajectory), AMD MI450 ramp evidence at Meta. - Late Feb 2027: NVDA Q4 FY27 and formal FY28 guide. Confirm the 71–72% margin trough, first look at Q1 FY28 guide (~$140B+ implied), and whether "supply-constrained 70%" has moved up (more HBM secured) or been reframed. - Mid-March: GTC 2027 – Rubin Ultra timing and an update to the $1T Blackwell+Rubin pipeline figure.
5. Customer capex and financing hold. Kress projected $1.3 trillion in capex from the top five hyperscalers next year and a cloud backlog of over $2 trillion. The non-hyperscale half is increasingly financed: with its customers' spending outpacing their cash generation, Nvidia is helping secure over $500 billion in capital from third parties. The company guarantees minimum revenue via take-or-pay contracts for part of the capacity, then earns from hardware sales and again from the rental upside. That works while capital is available; with Nvidia's financing model dependent on access to capital, rising long-term interest rates could tighten access for the labs and the neoclouds.
- The beat shrinks or the guide flattens. Nvidia has now beaten its own guidance for 13 consecutive quarters, but the margin of the beat has been narrowing – from a 22.8% revenue beat in Q2 FY2024 down to 4.6% last quarter, and 4.5% this quarter. A Q3 print near the $108B midpoint with a Q4 guide under ~$118–120B would mean the supply ramp is behind the curve needed for FY28. - Gross margin breaks below 71% or the FY28 recovery language softens. That tells you the +15% price increases aren't being fully absorbed – the first real crack in pricing power. - Purchase commitments stop growing / inventory builds. If the $279B figure stalls while inventory balloons, supply is arriving faster than customers can rack it (power/site readiness). The shortages span the artificial intelligence infrastructure provider's supply network, extending from high-speed memory and silicon vendors to suppliers of AI data centers under construction. - ACIE (neocloud/enterprise/sovereign) decelerates toward hyperscale growth. Last quarter non-hyperscale revenue grew 25% sequentially and 138% year-on-year versus hyperscale at 13%, and is on track to become the larger half of the segment. ACIE is the financing-dependent, less-visible half; if it stalls, so does the diversification story. - Custom silicon takes real share. Hyperscaler custom silicon programmes plus AMD ramp are gradually reducing NVIDIA share from approximately 70% in 2023 to approximately 55 to 60% in 2026. OpenAI published benchmarks for its custom Jalapeno chip, which topped Nvidia's Blackwell in efficiency, and Meta's AMD MI450 gigawatt deployment is in H2 2026. Any hyperscaler explicitly reallocating 2027 GPU budgets toward in-house silicon is a leading indicator. - Credit. Neocloud spreads widening, GPU-backed financings repricing, or a long bond stuck above 5% – the plumbing that funds the ACIE half.
The number: ~70% FY28 growth ≈ $680–700B revenue, and it's a supply ceiling, not a demand forecast. CFO Colette Kress said on a call with analysts that Nvidia expects fiscal 2028 revenue growth of 70%, while analysts were expecting 44%. Kress said that customer forecasts "point to our growth doubling next year," but she said guidance reflects supply constraints. Guidance for 70% revenue growth in fiscal 2028 implies around $690 billion ahead of a Street consensus estimate of $573.5 billion. Rough run-rate math: NVDA exits FY27 around $120B/quarter (Q3 guide $108B); $690B in FY28 needs an average of ~$170B/quarter, i.e. roughly 10–15% sequential growth every quarter for four more quarters. That's the bar. What has to go right 1. Memory (the binding constraint). Management forecast revenue growth of roughly 70% for fiscal 2028, noting that its production capacity was constrained by memory supply. NVDA has bought its way to the front of the line: purchase commitments to suppliers, including memory, more than doubled to $279 billion from $119 billion in the previous quarter. Those commitments must convert into delivered HBM4 on schedule from all three suppliers. 2. Packaging. HBM4 supply is constrained by advanced semiconductor packaging capacity at TSMC, and prices have risen sharply as AI infrastructure builders compete for the same supply. TSMC CoWoS confirmed at ~130,000 wpm by YE 2026 (up from ~75,000 in 2025, ~40,000 in 2024). FY28 needs another step-up beyond that. 3. Rubin ramp without a Blackwell-style hiccup. Vera Rubin GPU: production shipments Q3 FY27, volume ramp Q4 FY27. Vera Rubin is already in full production and ramping across CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Essentially all of FY28's incremental revenue is Rubin. 4. Pricing sticks, so margins trough and recover. Gross margins, while still exceptional at 75% for Q2, will compress to approximately 74% in Q3 and trough at 71–72% in Q4, before NVIDIA expects a recovery to 72–73% in fiscal 2028 as price increases in its supply agreements take effect. AI server prices will rise more than 15% on Vera Rubin and Grace Blackwell systems shipping in early 2027. Customers accepting +15% ASPs is what turns 70% unit-constrained growth into 70% revenue growth. 5. Customer capex and financing hold. Kress projected $1.3 trillion in capex from the top five hyperscalers next year and a cloud backlog of over $2 trillion. The non-hyperscale half is increasingly financed: with its customers' spending outpacing their cash generation, Nvidia is helping secure over $500 billion in capital from third parties. The company guarantees minimum revenue via take-or-pay contracts for part of the capacity, then earns from hardware sales and again from the rental upside. That works while capital is available; with Nvidia's financing model dependent on access to capital, rising long-term interest rates could tighten access for the labs and the neoclouds. Note China is not in the number – NVIDIA is not assuming any Data Center compute revenue from China in its outlook. That's upside optionality, not a dependency. Earliest signs it's slipping - The beat shrinks or the guide flattens. Nvidia has now beaten its own guidance for 13 consecutive quarters, but the margin of the beat has been narrowing – from a 22.8% revenue beat in Q2 FY2024 down to 4.6% last quarter, and 4.5% this quarter. A Q3 print near the $108B midpoint with a Q4 guide under ~$118–120B would mean the supply ramp is behind the curve needed for FY28. - Gross margin breaks below 71% or the FY28 recovery language softens. That tells you the +15% price increases aren't being fully absorbed – the first real crack in pricing power. - Purchase commitments stop growing / inventory builds. If the $279B figure stalls while inventory balloons, supply is arriving faster than customers can rack it (power/site readiness). The shortages span the artificial intelligence infrastructure provider's supply network, extending from high-speed memory and silicon vendors to suppliers of AI data centers under construction. - ACIE (neocloud/enterprise/sovereign) decelerates toward hyperscale growth. Last quarter non-hyperscale revenue grew 25% sequentially and 138% year-on-year versus hyperscale at 13%, and is on track to become the larger half of the segment. ACIE is the financing-dependent, less-visible half; if it stalls, so does the diversification story. - Custom silicon takes real share. Hyperscaler custom silicon programmes plus AMD ramp are gradually reducing NVIDIA share from approximately 70% in 2023 to approximately 55 to 60% in 2026. OpenAI published benchmarks for its custom Jalapeno chip, which topped Nvidia's Blackwell in efficiency, and Meta's AMD MI450 gigawatt deployment is in H2 2026. Any hyperscaler explicitly reallocating 2027 GPU budgets toward in-house silicon is a leading indicator. - Credit. Neocloud spreads widening, GPU-backed financings repricing, or a long bond stuck above 5% – the plumbing that funds the ACIE half. Calendar for the next two quarters - Late Sept: Micron FQ4 – HBM4 sold-out status and 2027 contract pricing; a rollover in memory prices would actually be good for NVDA margins but might signal cooling demand. - Mid-Oct: TSMC Q3 – CoWoS capacity adds for 2027; any capex cut is a red flag for FY28 supply. - Late Oct / early Nov: MSFT, GOOGL, AMZN, META Q3 – first 2027 capex framing. Benchmark it against Kress's $1.3T top-five figure; also watch commentary on custom accelerators vs GPUs. - ~Nov 19: NVDA Q3 FY27. Checklist: revenue vs $108B; Q4 guide (need ≥~$118B); GM ~74%; hyperscale vs ACIE split; purchase commitments; inventory; Rubin volume commentary; any change to the 70% frame. - Dec: Broadcom (custom ASIC trajectory), AMD MI450 ramp evidence at Meta. - Late Feb 2027: NVDA Q4 FY27 and formal FY28 guide. Confirm the 71–72% margin trough, first look at Q1 FY28 guide (~$140B+ implied), and whether "supply-constrained 70%" has moved up (more HBM secured) or been reframed. - Mid-March: GTC 2027 – Rubin Ultra timing and an update to the $1T Blackwell+Rubin pipeline figure. Bottom line: The 70% is a memory-and-packaging capacity number with demand running ~100%+, so FY28 breaks on supply execution and pricing stickiness first – watch the Q4 guide (≥$118B), GM holding the 71–72% floor, and hyperscaler 2027 capex framing in late October; demand erosion via custom silicon or neocloud credit stress is the slower, second-order risk.
Note China is not in the number – NVIDIA is not assuming any Data Center compute revenue from China in its outlook. That's upside optionality, not a dependency.
Short version: the ~[earnings recap agent]">70% guided FY2028 revenue growth is a supply number, not a demand number — so the things that have to go right are mostly on NVIDIA's own side (Vera Rubin ramp, memory supply, margins bottoming where promised), while the things that could break it are on the customer and financing side (hyperscaler capex, pre-profit AI labs, neocloud credit).
Management guided FY2028 revenue to grow ~[earnings recap agent]">70% year-over-year (guided), while stating unconstrained demand is closer to 100% and the [earnings recap agent]">70% reflects confirmed supply capacity [earnings recap]. Consensus is already there: FY2027 revenue ~$408.4B and FY2028 ~$685.9B (consensus), which works out to ~68% growth [estimate]. That means "holding" the outlook is roughly the base case, not a beat.
What has to go right
1. Rubin ships at record pace. Vera Rubin production shipments have begun with purchase orders from all major hyperscalers, and management expects the fastest ramp in its history [earnings recap]. The near-term proof point is Q3 FY2027 (quarter ending late October 2026): guided revenue [guidance]">$108.0B ±[guidance]">2% (guided) [guidance], up from a reported $96.2B in Q2 FY2027 (quarter ended 2026-07-26), itself +106% y/y and up from $81.6B in Q1 FY2027 [Q2 FY2027][Q1 FY2027]. Note the guide sat 0.6% below the $108.65B next-quarter consensus [earnings recap] — supply, not orders, capped it.
2. The supply chain NVIDIA just locked in actually delivers. Supply and capacity commitments jumped from $119B to $279B in one quarter (reported), with $92B due in the rest of FY2027 and [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$87B in the following year [10-Q notes]; total purchase and other commitments are [10-Q notes]">$366B [10-Q notes]. Inventories rose to a reported [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B the prior quarter and [Q2 FY2026]">$15.0B a year ago [Q2 FY2027][Q1 FY2027][Q2 FY2026]. For a fabless company that's the growth bet in physical form: [earnings recap agent]">70% only happens if TSMC, memory makers and ODMs convert that commitment into racks on schedule. The 10-Q still flags supply constraints as persisting [debate agent].
3. Memory inflation stops at the guided floor. Gross margin (GAAP) was a reported [state]">75.0% in Q2 FY2027, versus [Q1 FY2027]">74.9% the prior quarter and [Q2 FY2026]">72.4% a year ago [Q2 FY2027][Q1 FY2027][Q2 FY2026]. Guidance walks it down: [guidance]">74% ±50bp in Q3, a 71–72% bottom in Q4 FY2027 (quarter ending late January 2027), recovering to 72–73% for FY2028, all due to memory pricing that "exceeded prior expectations and are headed even higher" [earnings recap]. The 73% margin trigger is scheduled to fire on management's own guidance [watch]. Samsung and SK Hynix inventories have dropped below 10 days [web] — good for the demand thesis, bad for NVIDIA's cost of goods.
4. Customers keep funding it. The single highest-signal bear trigger is a coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL [watch]. Concentration is real: one direct customer was [guidance]">16% of Q2 revenue, three customers were [guidance]">16%, 15% and [10-Q notes]">13% of first-half revenue, and five customers hold [10-Q notes]">22%, [debate agent]">14%, [10-Q notes]">13%, [state]">11% and [10-Q notes]">10% of receivables (reported) [10-Q notes].
5. The financing layer holds. Management acknowledged nearly $50B of balance-sheet exposure to Frontier AI labs without investment-grade credit [earnings recap]. Non-marketable securities rose to a reported [10-Q notes]">$51.2B from [10-Q notes]">$22.3B [10-Q notes]; equity investment commitments are [Q2 FY2027]">$25B, AI cloud partnership commitments [Q3 FY2026]">$36B, cloud service agreements [10-Q notes]">$29B [10-Q notes]; and in August 2026 NVIDIA guaranteed up to $105B of SB Energy land/power/shell buildout (total guarantee exposure [10-Q notes]">$108.5B) [10-Q notes]. The reported ~$250B OpenAI compute/financing deal knocked the stock ~4% on 2026-07-27 [watch]. To fund all this, total debt went from a reported $8.5B to $33.4B after a [Q2 FY2027]">$25B June note issuance, while buybacks ran [Q2 FY2027]">$39.0B year-to-date [Q2 FY2027][10-Q notes].
Earliest signs it is slipping
Framing facts
Guidance excludes any China data-center compute revenue [guidance] — a policy thaw is upside not in the [earnings recap agent]">70%, and Huawei's push for domestic chips [web] makes a return less likely. The stock at [state]">$230.36 trades [state]">24.9x forward earnings (modeled/consensus) and prices ~[state]">29.9% annual FCF growth for a decade (modeled) [state]; peer-median P/S implies [state]">−69.5% [state], so a growth wobble hits the multiple, not just the estimate. Quality flags: Piotroski 4/9, accrual ratio [state]">10.9%, insiders net −4.69M shares with zero open-market buys [state].
Also disclosed, not material to this question: [Q2 FY2027]">$25B of data-center leases not yet commenced, $42.0B of debt securities in a short-duration unrealized-loss position, and the 38% non-US revenue mix [10-Q notes].
What to watch: (1) Q3 FY2027 revenue vs the [guidance]">$108.0B guide and whether Q4 guidance implies FY2028 still on a ~[earnings recap agent]">70% path; (2) Q3 gross margin vs [guidance]">74% and any change to the 71–72% Q4 floor; (3) hyperscaler 2027 capex commentary at the October/November earnings round, plus DSO and the OpenAI deal structure.
Bottom line: [earnings recap agent]">70% is achievable because it's already the supply-limited number and customers are prepaying for it; the risk isn't that demand is fake, it's that the margin floor gives way or the customers and labs NVIDIA is now financing can't keep paying.
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
Short version: the ~70% guided FY2028 revenue growth is a supply number, not a demand number – so the things that have to go right are mostly on NVIDIA's own side (Vera Rubin ramp, memory supply, margins bottoming where promised), while the things that could break it are on the customer and financing side (hyperscaler capex, pre-profit AI labs, neocloud credit).
Management guided FY2028 revenue to grow ~70% year-over-year (guided), while stating unconstrained demand is closer to 100% and the 70% reflects confirmed supply capacity . Consensus is already there: FY2027 revenue ~$408.4B and FY2028 ~$685.9B (consensus), which works out to ~68% growth . That means "holding" the outlook is roughly the base case, not a beat.
2. The supply chain NVIDIA just locked in actually delivers. Supply and capacity commitments jumped from $119B to $279B in one quarter (reported), with $92B due in the rest of FY2027 and $87B in the following year ; total purchase and other commitments are $366B . Inventories rose to a reported $31.6B from $25.8B the prior quarter and $15.0B a year ago Q2 FY2027Q1 FY2027Q2 FY2026. For a fabless company that's the growth bet in physical form: 70% only happens if TSMC, memory makers and ODMs convert that commitment into racks on schedule. The 10-Q still flags supply constraints as persisting .
Short version: the ~70% guided FY2028 revenue growth is a supply number, not a demand number – so the things that have to go right are mostly on NVIDIA's own side (Vera Rubin ramp, memory supply, margins bottoming where promised), while the things that could break it are on the customer and financing side (hyperscaler capex, pre-profit AI labs, neocloud credit).
Short version: the ~70% guided FY2028 revenue growth is a supply number, not a demand number – so the things that have to go right are mostly on NVIDIA's own side (Vera Rubin ramp, memory supply, margins bottoming where promised), while the things that could break it are on the customer and financing side (hyperscaler capex, pre-profit AI labs, neocloud credit).
4. Customers keep funding it. The single highest-signal bear trigger is a coordinated 2027 capex-guidance cut from MSFT/AMZN/META/GOOGL . Concentration is real: one direct customer was 16% of Q2 revenue, three customers were 16%, 15% and 13% of first-half revenue, and five customers hold 22%, 14%, 13%, 11% and 10% of receivables (reported) .
5. The financing layer holds. Management acknowledged nearly $50B of balance-sheet exposure to Frontier AI labs without investment-grade credit . Non-marketable securities rose to a reported $51.2B from $22.3B ; equity investment commitments are $25B, AI cloud partnership commitments $36B, cloud service agreements $29B ; and in August 2026 NVIDIA guaranteed up to $105B of SB Energy land/power/shell buildout (total guarantee exposure $108.5B) . The reported ~$250B OpenAI compute/financing deal knocked the stock ~4% on 2026-07-27 . To fund all this, total debt went from a reported $8.5B to $33.4B after a $25B June note issuance, while buybacks ran $39.0B year-to-date Q2 FY2027 .
Short version: the ~70% guided FY2028 revenue growth is a supply number, not a demand number – so the things that have to go right are mostly on NVIDIA's own side (Vera Rubin ramp, memory supply, margins bottoming where promised), while the things that could break it are on the customer and financing side (hyperscaler capex, pre-profit AI labs, neocloud credit).
- Receivables outrunning revenue. DSO stretched to ~60 days in Q2 FY2027 from ~45 the prior quarter and ~54 a year ago, with receivables at a reported $63.1B Q2 FY2027Q1 FY2027Q2 FY2026. Another leg up would suggest customers are being carried. Offsetting evidence for now: customer advances of $15.6B in the first half vs $7.5B a year earlier – people are prepaying. - Margin undershoot. A Q3 print below 73.5% or a Q4 guide below 71% (absent a China charge) would say memory is eating more than modeled . - Neocloud credit. CoreWeave CDS recovered to ~452bps from ~881bps; back above 600bps signals renewed stress . - Roadmap slip. A confirmed Kyber/Rubin Ultra delay that lets AMD MI450 or Google Ironwood take 2027–28 rack slots ; the real competitive tell is an absolute-dollar inference crossover to ASICs, not a share statistic . - Estimate direction. Revisions are currently 5 up / 0 down, with FY2028 consensus EPS nudged to $15.64 ; the first coordinated cuts would matter more than any headline.
Guidance excludes any China data-center compute revenue – a policy thaw is upside not in the 70%, and Huawei's push for domestic chips makes a return less likely. The stock at $230.36 trades 24.9x forward earnings (modeled/consensus) and prices ~29.9% annual FCF growth for a decade (modeled) ; peer-median P/S implies −69.5% , so a growth wobble hits the multiple, not just the estimate. Quality flags: Piotroski 4/9, accrual ratio 10.9%, insiders net −4.69M shares with zero open-market buys .
The ~[earnings recap agent]">70% FY2028 outlook is a supply number, not a demand number — so the things that have to go right are mostly about shipping and margins, and the earliest slip signals will show up in the balance sheet and at customers before they show up in revenue.
Why [earnings recap agent]">70% is the bar, and what it needs mathematically
Management guided FY28 revenue up ~[earnings recap agent]">70% while saying true demand is growing closer to 100%, with the gap being confirmed supply capacity [earnings recap]. Consensus already sits at that bar: FY2027 revenue ~$408.4B and FY2028 ~$685.9B [estimate], i.e. +68%. Working from the Q3 FY2027 guide of [guidance]">$108.0B ±[guidance]">2% [guidance] and the FY2027 consensus, Q4 FY2027 is implied at roughly $122–123B [derived]; getting from that exit rate to ~$686B requires sequential growth of roughly 13–14% every quarter through FY2028 [derived]. For reference, the Q2→Q3 step is ~12% (from $96.2B [Q2 FY2027] to [guidance]">$108.0B [guidance]). So the bar is "keep doing what you just did, for six more quarters, on a bigger base."
What has to go right — four legs
Vera Rubin ramps on schedule and supply commitments convert to units. Vera Rubin began production shipments with purchase orders from every major hyperscaler [earnings recap]. The 10-Q shows the company pre-committed for it: supply and capacity commitments jumped from $119B to $279B in one quarter, with $92B due the rest of this fiscal year and [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$87B/[Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$88B in the next two [10-Q notes]; inventory rose to [Q2 FY2027]">$31.6B from [Q1 FY2027]">$25.8B [Q2 FY2027, Q1 FY2027]. HBM4 is the pinch point: Samsung reportedly reached ~80% HBM4 yield in late August [web?] — a single-source-style claim, treat as unconfirmed. Kyber/Rubin Ultra timing is on watch as a roadmap risk [watch].
Hyperscaler 2027 capex holds — the single highest-signal trigger [watch]. The web tape is currently supportive: management commentary suggests AI infrastructure spending will remain elevated into 2027, with little indication on the earnings calls that investment is nearing a peak [web], and Alphabet flagged a further 2027 increase while Meta and Microsoft have declined to provide specific guidance [web]. But FY26 free cash flows are expected to move close to zero or turn negative for all except Alphabet and Microsoft [web], and Alphabet's shares fell ~7% after raising 2026 capex in late July [web] — investor tolerance for the spend is thinning even as the spend continues.
Gross margin bottoms where guided and no lower. The path is [guidance]">74% ±50bp in Q3, a 71–72% trough in Q4 FY2027, recovering to 72–73% for FY28, driven by memory prices that exceeded expectations and are "headed even higher" [earnings recap]. The gm_slip trigger at 73% is scheduled to fire on management's own Q4 guide [watch]. Third-party rack cost work (Morgan Stanley, Bernstein via press) puts memory at ~26–29% of a Vera Rubin rack's bill of materials versus ~[watch]">9% for GB300 [web?] — estimates, not disclosed figures. Because Nvidia is fabless, memory inflation hits its own COGS only to the extent it can't reprice; the guide implies partial pass-through.
Customer financing doesn't crack. Management acknowledged nearly $50B of exposure to pre-profit frontier labs [earnings recap]; non-marketable securities doubled to [10-Q notes]">$51.2B from [10-Q notes]">$22.3B [10-Q notes]; guarantees capped at $105B were signed in August with SB Energy affiliates for land, power and shell buildouts [10-Q notes]; and [Q3 FY2026]">$36B of AI-cloud partnership plus [10-Q notes]">$29B of cloud-service commitments sit alongside [10-Q notes]. Total future purchase and other commitments are [10-Q notes]">$366B [10-Q notes]. The company also moved from $8.5B to $33.4B debt via a [Q2 FY2027]">$25B June note issue [10-Q notes, Q2 FY2027] while buying back [Q2 FY2027]">$39B YTD [Q2 FY2027]. The [earnings recap agent]">70% number assumes this ecosystem keeps paying.
Earliest signs it is slipping
What to watch over the next two quarters
Late Oct–early Nov: MSFT/AMZN/META/GOOGL September-quarter calls — does any of them put a number on 2027 capex, and is it up [web]. Mid/late Nov: Nvidia Q3 FY2027 print — revenue vs [guidance]">$108B, GAAP GM vs [guidance]">74%, opex vs ~[guidance]">$9.2B [guidance], DSO, advances, and the Q4 revenue/GM guide. Ongoing: memory spot/contract pricing and HBM4 yield confirmations from SK Hynix/Samsung/Micron; any confirmation, denial or structure detail on the reported OpenAI financing deal [watch]; China export-policy moves — guidance excludes China data-center compute, so a thaw is pure upside and a tightening removes only optionality [guidance, watch]; antitrust developments including the China finding [debate].
Also disclosed, not material to this question: $42.0B of debt securities in a continuous unrealized-loss position (losses not significant); $36.9B of restricted publicly held equity securities; [Q2 FY2027]">$25B of data-center leases not yet commenced; 38% of Q2 revenue from customers headquartered outside the U.S. [10-Q notes].
Bottom line: The [earnings recap agent]">70% is achievable only if Vera Rubin ships against [Supply And Capacity Commitments] (as of 2026-07-26) = total $279.0B; rest of FY $92.0B; yr 1 $87.0B; yr 2 $88.0B; yr 3 $6.0B; yr 4 $5.0B; after $1.0B — table row “Supply and capacity” [10-Q notes]">$279B of supply commitments, memory-driven margin compression stops at 71–72%, and hyperscalers plus vendor-financed labs keep paying — watch DSO, customer advances and the Q4 guide in November for the first crack, and the hyperscalers' 2027 capex language a few weeks earlier.
Sources (most reliable first): 2026 hyperscaler capex tops US$700bn – analysis | TMT Finance [src:unverified] · Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow [src:unverified]
The rule matched these words in context. This is not independent verification or a reasoning score. “Not matched” can mean absent content or wording the rule did not recognize.
The ~70% FY2028 outlook is a supply number, not a demand number – so the things that have to go right are mostly about shipping and margins, and the earliest slip signals will show up in the balance sheet and at customers before they show up in revenue.
Management guided FY28 revenue up ~70% while saying true demand is growing closer to 100%, with the gap being confirmed supply capacity . Consensus already sits at that bar: FY2027 revenue ~$408.4B and FY2028 ~$685.9B , i.e. +68%. Working from the Q3 FY2027 guide of $108.0B ±2% and the FY2027 consensus, Q4 FY2027 is implied at roughly $122–123B ; getting from that exit rate to ~$686B requires sequential growth of roughly 13–14% every quarter through FY2028 . For reference, the Q2→Q3 step is ~12% (from $96.2B Q2 FY2027 to $108.0B ). So the bar is "keep doing what you just did, for six more quarters, on a bigger base."
1. Vera Rubin ramps on schedule and supply commitments convert to units. Vera Rubin began production shipments with purchase orders from every major hyperscaler . The 10-Q shows the company pre-committed for it: supply and capacity commitments jumped from $119B to $279B in one quarter, with $92B due the rest of this fiscal year and $87B/$88B in the next two ; inventory rose to $31.6B from $25.8B . HBM4 is the pinch point: Samsung reportedly reached ~80% HBM4 yield in late August – a single-source-style claim, treat as unconfirmed. Kyber/Rubin Ultra timing is on watch as a roadmap risk .
3. Gross margin bottoms where guided and no lower. The path is 74% ±50bp in Q3, a 71–72% trough in Q4 FY2027, recovering to 72–73% for FY28, driven by memory prices that exceeded expectations and are "headed even higher" . The gmslip trigger at 73% is scheduled to fire on management's own Q4 guide . Third-party rack cost work (Morgan Stanley, Bernstein via press) puts memory at ~26–29% of a Vera Rubin rack's bill of materials versus ~9% for GB300 – estimates, not disclosed figures. Because Nvidia is fabless, memory inflation hits its own COGS only to the extent it can't reprice; the guide implies partial pass-through.
1. Vera Rubin ramps on schedule and supply commitments convert to units. Vera Rubin began production shipments with purchase orders from every major hyperscaler . The 10-Q shows the company pre-committed for it: supply and capacity commitments jumped from $119B to $279B in one quarter, with $92B due the rest of this fiscal year and $87B/$88B in the next two ; inventory rose to $31.6B from $25.8B . HBM4 is the pinch point: Samsung reportedly reached ~80% HBM4 yield in late August – a single-source-style claim, treat as unconfirmed. Kyber/Rubin Ultra timing is on watch as a roadmap risk .
2. Hyperscaler 2027 capex holds – the single highest-signal trigger . The web tape is currently supportive: management commentary suggests AI infrastructure spending will remain elevated into 2027, with little indication on the earnings calls that investment is nearing a peak , and Alphabet flagged a further 2027 increase while Meta and Microsoft have declined to provide specific guidance . But FY26 free cash flows are expected to move close to zero or turn negative for all except Alphabet and Microsoft , and Alphabet's shares fell ~7% after raising 2026 capex in late July – investor tolerance for the spend is thinning even as the spend continues.
4. Customer financing doesn't crack. Management acknowledged nearly $50B of exposure to pre-profit frontier labs ; non-marketable securities doubled to $51.2B from $22.3B ; guarantees capped at $105B were signed in August with SB Energy affiliates for land, power and shell buildouts ; and $36B of AI-cloud partnership plus $29B of cloud-service commitments sit alongside . Total future purchase and other commitments are $366B . The company also moved from $8.5B to $33.4B debt via a $25B June note issue while buying back $39B YTD Q2 FY2027. The 70% number assumes this ecosystem keeps paying.
- Receivables outrunning revenue. DSO stretched to ~60 days from ~45 in Q1 , with receivables at $63.1B and five customers holding 22/14/13/11/10% of the balance . Another 10-day rise in Q3 with flat cash conversion is the first hard-data warning. - Customer advances stall. Deferred-revenue additions included $15.6B of customer advances in H1 vs $7.5B a year ago ; a sequential drop means customers are no longer pre-paying to reserve supply. - Q3 lands below $108B or Q4 is guided below ~$120B – that breaks the 13–14% sequential cadence . - Gross margin below the 73.5% low end in Q3 or a Q4 guide below 71% – memory is winning the pricing fight. - Neocloud credit. CoreWeave CDS recovered from ~881bp to ~452bp; back above 600bp signals renewed stress . - Concentration. One customer was 16% of Q2 revenue ; a top customer publicly leaning on its own silicon is the inference-share crossover to watch .
- Receivables outrunning revenue. DSO stretched to ~60 days from ~45 in Q1 , with receivables at $63.1B and five customers holding 22/14/13/11/10% of the balance . Another 10-day rise in Q3 with flat cash conversion is the first hard-data warning. - Customer advances stall. Deferred-revenue additions included $15.6B of customer advances in H1 vs $7.5B a year ago ; a sequential drop means customers are no longer pre-paying to reserve supply. - Q3 lands below $108B or Q4 is guided below ~$120B – that breaks the 13–14% sequential cadence . - Gross margin below the 73.5% low end in Q3 or a Q4 guide below 71% – memory is winning the pricing fight. - Neocloud credit. CoreWeave CDS recovered from ~881bp to ~452bp; back above 600bp signals renewed stress . - Concentration. One customer was 16% of Q2 revenue ; a top customer publicly leaning on its own silicon is the inference-share crossover to watch .
Late Oct–early Nov: MSFT/AMZN/META/GOOGL September-quarter calls – does any of them put a number on 2027 capex, and is it up . Mid/late Nov: Nvidia Q3 FY2027 print – revenue vs $108B, GAAP GM vs 74%, opex vs ~$9.2B , DSO, advances, and the Q4 revenue/GM guide. Ongoing: memory spot/contract pricing and HBM4 yield confirmations from SK Hynix/Samsung/Micron; any confirmation, denial or structure detail on the reported OpenAI financing deal ; China export-policy moves – guidance excludes China data-center compute, so a thaw is pure upside and a tightening removes only optionality ; antitrust developments including the China finding .
Checklist coverage is a contextual lexical check with retained passages, not an independent accuracy score. Numeric matches identify scalar occurrences in saved evidence; they do not validate claims. Reference tags are not unique sources. Compare recorded cost separately. Earlier reviewer notes are archived in legacy_review in the source JSON.
Pick a model. The table shows the same 9 questions from memory, with web search, with Finbly’s research, and with Finbly’s research plus web search, wherever that run exists. Web search and Finbly + web were run on the Claude models and GPT only.
| Across 9 questions | No research | Web search | Finbly research | Finbly + web |
|---|---|---|---|---|
| Checklist matchescontextual text matches, with the supporting passage retained; not factual accuracy | 30% | 51% | 85% | 82% |
| Numbers in saved evidencescalar occurrences, not verification of the associated claims | 104 | 191 | 371 | 304 |
| Claim accuracynot assessed by the numeric occurrence checker | Not assessed | Not assessed | Not assessed | Not assessed |
| Numbers without a scalar matchno scalar match in the saved packet, or an ambiguous range; not an error count | 192 | 295 | 70 | 107 |
| Research reference tagscategory labels in the answer; repeats counted | 0 | 0 | 231 | 249 |
| Input tokens per questionwhat the model read | 130 | 46.3k | 11.0k | 27.1k |
| Cost per questionmodel list prices; a dash means no published rate | $0.14 | $0.65 | $0.22 | $0.36 |
Key-item coverage and selected figures are checked against Finbly’s stores; costs use model list prices. These checks do not measure the quality of reasoning or establish that one model is better overall.
“Looking ahead to FY2028: what has to go right for NVIDIA's ~70% growth outlook to hold, what are the earliest signs it is slipping, and what should I watch over the next two quarters?” asked under the available research setups. Checklist matches measure answer content; the separate forward-call ledger tracks risk conditions and calls requiring reader review.
| One question | Web search | Finbly research | Finbly + web |
|---|---|---|---|
| Checklist matchescontextual text matches for the forward checklist; not forecast accuracy | 82% | 91% | 91% |
| Numbers in saved evidencescalar occurrences in the saved packet; not claim verification | 27 | 63 | 43 |
| Claim accuracynot assessed by the numeric occurrence checker | Not assessed | Not assessed | Not assessed |
| Figures not verifiedthe check found no source either way | 28 | 13 | 20 |
| Referencesevidence tags a reader can follow | 0 | 43 | 38 |
| Web searches usedweb-enabled answers may choose not to search | 4 | 0 | 2 |
| Input tokenswhat the model read, including search results and cached reads | 67.3k | 14.9k | 52.4k |
| Cost · timemodel list prices; wall-clock seconds | $0.99 · 79s | $0.39 · 62s | $0.77 · 61s |
| Forward calls loggedcheckable claims the answer made, from the ledger below the forward-looking scenario | 6 | 7 | 7 |
Run on 6 models; Kimi and Qwen have no native web search, so their first column is the model from memory. Checklist matches measure content, not forecast accuracy. Forward-call conditions and reader-review notes are shown separately in the ledger.
Runs are ordered by recorded estimated cost. Use the checks alongside the original answers; a lower cost or more matches does not establish better research.
| Research setup | Key items | Numbers in saved evidence | Claim accuracy | Not verified | Reference tags | Tokens in | Tokens out | Cost | Per question |
|---|---|---|---|---|---|---|---|---|---|
| Kimi K2.6 + Finbly research9 questions | 82%69 of 84 | 260 | Not assessed | 52 | 181 | 84.9k | 10.8k | $0.095 | $0.011 |
| Kimi K3 + Finbly research9 questions | 90%76 of 84 | 291 | Not assessed | 40 | 154 | 85.7k | 10.5k | $0.39 | $0.043 |
| Claude Sonnet 5 + Finbly research9 questions | 80%67 of 84 | 197 | Not assessed | 42 | 128 | 98.9k | 23.8k | $0.57 | $0.063 |
| GPT-5.6 Sol + Finbly research9 questions | 63%53 of 84 | 262 | Not assessed | 80 | 180 | 65.6k | 10.4k | $0.59 | $0.065 |
| Claude Sonnet 5 + Finbly research + web search9 questions | 75%63 of 84 | 146 | Not assessed | 45 | 100 | 153.6k | 22.6k | $0.68 | $0.075 |
| Claude Sonnet 5 + web search, no Finbly9 questions | 38%32 of 84 | 111 | Not assessed | 172 | 0 | 373.4k | 20.4k | $1.71 | $0.19 |
| Claude Fable 5.1 + Finbly research9 questions | 85%71 of 84 | 371 | Not assessed | 70 | 231 | 98.9k | 26.5k | $2.02 | $0.22 |
| GPT-5.6 Sol + web search, no Finbly9 questions | 49%41 of 84 | 126 | Not assessed | 156 | 0 | 286.7k | 15.1k | $2.08 | $0.23 |
| Claude Fable 5.1 + Finbly research + web search9 questions | 82%69 of 84 | 304 | Not assessed | 107 | 249 | 243.9k | 24.5k | $3.28 | $0.36 |
| Claude Fable 5.1 + web search, no Finbly9 questions | 51%43 of 84 | 191 | Not assessed | 295 | 0 | 416.7k | 28.4k | $5.89 | $0.65 |
| Qwen 3.8 Max + Finbly research9 questions | 79%66 of 84 | 302 | Not assessed | 36 | 206 | 95.5k | 12.8k | — | — |
Recorded input volume. Finbly research runs used 7.3k to 11.0k input tokens per question; web-search runs used 31.9k to 46.3k. Runs without research used 92 to 180; the counted full-document bundle contains 313.3k tokens. These are different input setups, not a measurement of subscription savings.
Key items measures mentions of reviewer-selected facts, not whether the reasoning is right. Input tokens include cached reads; output tokens count what the model wrote. Costs use model list prices; a dash means no published rate. Building and refreshing Finbly’s research is excluded. Different runs may contain different numbers of questions; inspect the per-question costs and individual answers.
Each model’s answers are plotted against the quarters Finbly had on file when the comparison ran. Below the timeline: the source bundle behind the answers and what sending it whole would cost.
5 SEC filings, 9 earnings calls and 134 headlines, plus 1,129 verified financial facts across 5 fiscal years. The full-document comparison bundle contains about 288 estimated pages. Finbly selects evidence for each question.
Bundle counted 8 Sept 2026 using Anthropic count_tokens (claude-sonnet-5). This is a calculated scenario: repeatedly sending the document bundle at the selected model’s stored list prices, compared with recorded Finbly answer costs. It is separate from the answer-quality experiment above. It excludes building and maintaining the research, subscription fees and customer time; the cached scenario assumes a 90% input-cache discount.
| Document | Filed | Pages | Tokens |
|---|---|---|---|
| filing10-K · period ended 2026-01-25 | 2026-02-25 | 114 | 120.7k |
| filing10-Q · period ended 2026-07-26 | 2026-08-26 | 51 | 66.1k |
| filing10-Q · period ended 2026-04-26 | 2026-05-20 | 46 | 57.6k |
| transcript4 earnings calls · Q3 FY2026, Q4 FY2026, Q1 FY2027, Q2 FY2027 | 2026-08-26 | 73 | 62.5k |
| news126 headlines, last 30 days | 2026-09-08 | 4 | 6.4k |
How the selected model answered a fixed set of company questions under different research setups. Original answers, recorded usage and selected numeric checks are available for inspection.
An overall accuracy rate, better investment returns, or superiority to the full ChatGPT or Claude research experience. Checking against Finbly’s own evidence is not independent validation.
Saved comparison: 8 Sept 2026. Company evidence snapshot: 8 Sept 2026. Today in Singapore: 12 Sept 2026. These dates change only when the comparison is regenerated.
| Model and setup | Baseline (see setup) | Finbly (see setup) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Research tags | Match | Claim accuracy | Not verified | Cost | Research tags | Match | Claim accuracy | Not verified | Cost | |
| 9 qClaude Fable 5.1 | 0 | 104/296 | Not assessed | 192 | $1.22 | 231 | 371/441 | Not assessed | 70 | $2.02 |
| 9 qGPT-5.6 Sol | 0 | 39/109 | Not assessed | 70 | $0.22 | 180 | 262/342 | Not assessed | 80 | $0.59 |
| 9 qClaude Sonnet 5 | 0 | 19/78 | Not assessed | 59 | $0.13 | 128 | 197/239 | Not assessed | 42 | $0.57 |
| 9 qQwen 3.8 Max | 0 | 31/122 | Not assessed | 91 | — | 206 | 302/338 | Not assessed | 36 | — |
| 9 qKimi K3 | 0 | 41/134 | Not assessed | 93 | $0.087 | 154 | 291/331 | Not assessed | 40 | $0.39 |
| 9 qKimi K2.6 | 0 | 56/174 | Not assessed | 118 | $0.020 | 181 | 260/312 | Not assessed | 52 | $0.095 |
| 9 qClaude Fable 5.1 · Finbly + web | 0 | 191/486 | Not assessed | 295 | $5.89 | 249 | 304/411 | Not assessed | 107 | $3.28 |
| 9 qClaude Fable 5.1 · web search only | 0 | 191/486 | Not assessed | 295 | $5.89 | 231 | 371/441 | Not assessed | 70 | $2.02 |
| 9 qClaude Sonnet 5 · Finbly + web | 0 | 111/283 | Not assessed | 172 | $1.71 | 100 | 146/191 | Not assessed | 45 | $0.68 |
| 9 qClaude Sonnet 5 · web search only | 0 | 111/283 | Not assessed | 172 | $1.71 | 128 | 197/239 | Not assessed | 42 | $0.57 |
| 9 qGPT-5.6 Sol · web search only | 0 | 126/282 | Not assessed | 156 | $2.08 | 180 | 262/342 | Not assessed | 80 | $0.59 |
Scoring v2 uses the original saved evidence, not today’s metrics. Signed scalar values must match exactly or round to the precision stated in the answer. A matching number does not verify its entity, metric, period or interpretation; ambiguous ranges stay unverified. Claim accuracy is not assessed. Checklist matches require words and context from the same passage, with explicit exceptions for multi-paragraph calendar coverage; expand each answer’s checklist to inspect its evidence. Reference tags are counted separately from source links; missing legacy web-citation metadata cannot be reconstructed. Earlier accuracy rankings have been withdrawn. The selected baseline used no supplied evidence and no browsing; the other side used Finbly evidence without browsing.
Explore NVIDIA’s filings, financial history and dated specialist research.
Open NVIDIA research