Original saved answers, research setups and evaluation details. Open a question or a model to examine the results for yourself.
K3 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 · 21 and 76 of 84 | 25% | 90% |
| Numbers in saved evidencescalar occurrences only; not claim verification · of 134 and 331 checked | 41 | 291 |
| 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 | 93 | 40 |
| Research reference tagscategory labels in the answer; repeats counted | 0 | 154 |
| Input tokens per questionwhat the model read | 180 | 9.5k |
| Cost per questionmodel list prices; a dash means no published rate | $0.0097 | $0.043 |
Saved model outputs are preserved. Underlines describe automated numeric checks; evidence tags identify research categories and are not direct source citations.
Yes — still very strong, but compressing modestly at the margin.
The numbers (GAAP, latest reported quarters):
Why it's drifting down:
Context: 73–75% gross margin on ~$35B+ quarterly revenue is still best-in-class for semis and well above AMD (~53%) or Intel. The margin trajectory is a deliberate trade — NVDA is sacrificing ~300–400 bps near-term to lock in the rack-scale architecture cycle.
What to watch: Q4 print vs. the 73% guide, and whether management reaffirms the "mid-70s as Blackwell scales" path for FY2026. A guide below [Q2 FY2026]">72% would change the story.
Bottom line: Margin is holding at elite levels; the dip to ~73% is a planned Blackwell-ramp cost, not erosion — watch the FY2026 recovery back to mid-70s.
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.
Context: 73–75% gross margin on ~$35B+ quarterly revenue is still best-in-class for semis and well above AMD (~53%) or Intel. The margin trajectory is a deliberate trade – NVDA is sacrificing ~300–400 bps near-term to lock in the rack-scale architecture cycle.
What to watch: Q4 print vs. the 73% guide, and whether management reaffirms the "mid-70s as Blackwell scales" path for FY2026. A guide below 72% would change the story.
Holding for now — [state]">75.0% reported in Q2 FY2027 — but management has already told you it won't: guidance resets margins to [guidance]">74% next quarter, then a 71–72% trough in Q4 FY2027.
The number and its basis. NVIDIA printed [state]">75.0% GAAP gross margin in Q2 FY2027 (quarter ended July 2026) [Q2 FY2027] — [F]; **Q2 FY2027 printed 75.0% GAAP** (gross profit $72.1B on $96.2B) [F, 10-Q]. The "margin compression" a screen sees is a transient China artifact, not structural — the *guided* compression into Q4 FY2027 (71–72%) is a memory-cost stor">$72.1B of gross profit on $96.2B revenue. That is the peak of the recent range, not a floor.
The trend. Against the prior quarter it's flat-to-up: Q1 FY2027 (quarter ended April 2026) printed [Q1 FY2027]">74.9% [Q1 FY2027]. Against the year-ago quarter it looks dramatically better — [Q2 FY2026]">72.4% in Q2 FY2026 (quarter ended July 2025), and Q1 FY2026 was [Q1 FY2026]">60.5% [Q1 FY2026][Q2 FY2026] — but that dip was a one-time ~$4.5B H20/China write-down that depressed all of FY2026 to [state]">71.1% (ex-charge ~73%) [state]. So the year-ago comparison flatters; the real run-rate has been 73–75% for four straight quarters.
Why it's about to compress. This is a guided, structural step-down, not a scare. NVIDIA is fabless, so memory-cost inflation hits its bill of materials directly: the CFO said memory (HBM/DRAM) price increases "exceeded prior expectations and are headed even higher," forcing GAAP margins to [guidance]">74.0% ±50bp in Q3 FY2027 (guided), a 71–72% trough in Q4 FY2027, and 72–73% for FY2028 [guidance] [earnings recap agent]. Management framed it as manageable, but it is a margin reset by their own admission. My watch flag on this is scheduled to fire on management's own guidance, since 71–72% sits below the 73% "pricing power intact" threshold [watch].
What frames it.
What to watch: (1) the Q3 print (earnings 2026-11-18) — does [guidance]">74.0% ±50bp hold, and does the Q4 71–72% trough guidance stand or deepen; (2) DRAM/HBM contract pricing into 2027 — if memory inflation outruns the guided reset, the 72–73% FY2028 recovery is at risk; (3) whether the stock treats a guided margin dip as noise, as it did with the [earnings recap agent]">+8.7% post-earnings move [earnings recap agent].
Bottom line: The margin is holding today at [state]">75.0%, but the honest answer is that it's guided down ~300–400bp by Q4 on memory costs — the question isn't whether it slips, it's whether 71–72% is the real bottom.
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.
Holding for now – 75% reported in Q2 FY2027 – but management has already told you it won't: guidance resets margins to 74% next quarter, then a 71–72% trough in Q4 FY2027.
The trend. Against the prior quarter it's flat-to-up: Q1 FY2027 (quarter ended April 2026) printed 74.9% Q1 FY2027. Against the year-ago quarter it looks dramatically better – 72.4% in Q2 FY2026 (quarter ended July 2025), and Q1 FY2026 was 60.5% Q1 FY2026Q2 FY2026 – but that dip was a one-time ~$4.5B H20/China write-down that depressed all of FY2026 to 71.1% (ex-charge ~73%) . So the year-ago comparison flatters; the real run-rate has been 73–75% for four straight quarters.
The trend. Against the prior quarter it's flat-to-up: Q1 FY2027 (quarter ended April 2026) printed 74.9% Q1 FY2027. Against the year-ago quarter it looks dramatically better – 72.4% in Q2 FY2026 (quarter ended July 2025), and Q1 FY2026 was 60.5% Q1 FY2026Q2 FY2026 – but that dip was a one-time ~$4.5B H20/China write-down that depressed all of FY2026 to 71.1% (ex-charge ~73%) . So the year-ago comparison flatters; the real run-rate has been 73–75% for four straight quarters.
Why it's about to compress. This is a guided, structural step-down, not a scare. NVIDIA is fabless, so memory-cost inflation hits its bill of materials directly: the CFO said memory (HBM/DRAM) price increases "exceeded prior expectations and are headed even higher," forcing GAAP margins to 74% ±50bp in Q3 FY2027 (guided), a 71–72% trough in Q4 FY2027, and 72–73% for FY2028 . Management framed it as manageable, but it is a margin reset by their own admission. My watch flag on this is scheduled to fire on management's own guidance, since 71–72% sits below the 73% "pricing power intact" threshold .
Holding for now – 75% reported in Q2 FY2027 – but management has already told you it won't: guidance resets margins to 74% next quarter, then a 71–72% trough in Q4 FY2027.
Why it's about to compress. This is a guided, structural step-down, not a scare. NVIDIA is fabless, so memory-cost inflation hits its bill of materials directly: the CFO said memory (HBM/DRAM) price increases "exceeded prior expectations and are headed even higher," forcing GAAP margins to 74% ±50bp in Q3 FY2027 (guided), a 71–72% trough in Q4 FY2027, and 72–73% for FY2028 . Management framed it as manageable, but it is a margin reset by their own admission. My watch flag on this is scheduled to fire on management's own guidance, since 71–72% sits below the 73% "pricing power intact" threshold .
Why it's about to compress. This is a guided, structural step-down, not a scare. NVIDIA is fabless, so memory-cost inflation hits its bill of materials directly: the CFO said memory (HBM/DRAM) price increases "exceeded prior expectations and are headed even higher," forcing GAAP margins to 74% ±50bp in Q3 FY2027 (guided), a 71–72% trough in Q4 FY2027, and 72–73% for FY2028 . Management framed it as manageable, but it is a margin reset by their own admission. My watch flag on this is scheduled to fire on management's own guidance, since 71–72% sits below the 73% "pricing power intact" threshold .
Why it's about to compress. This is a guided, structural step-down, not a scare. NVIDIA is fabless, so memory-cost inflation hits its bill of materials directly: the CFO said memory (HBM/DRAM) price increases "exceeded prior expectations and are headed even higher," forcing GAAP margins to 74% ±50bp in Q3 FY2027 (guided), a 71–72% trough in Q4 FY2027, and 72–73% for FY2028 . Management framed it as manageable, but it is a margin reset by their own admission. My watch flag on this is scheduled to fire on management's own guidance, since 71–72% sits below the 73% "pricing power intact" threshold .
Price targets:
Valuation:
What's embedded in estimates:
Key risk to the numbers: estimates have been revised upward almost every quarter for two years; any signal of capex digestion among hyperscalers or China/export restrictions tightening would compress both the EPS estimate and the multiple simultaneously.
Bottom line: consensus target ~$175 with ~[state]">30% implied upside, and a forward P/E near 30x that looks cheap on PEG only if the [F] — unresolved. AMD's **Helios/MI455X** (Advancing AI, July 22–23) is a legitimate spec-sheet win on memory (31 TB HBM4, ~50% more) but is priced **~40% above** Vera Rubin ($5.0–5.5M vs $3.5–4.0M/rack [E]) and rides dilutive OpenAI/Meta warrants; Google **Ironwood** claims ~44% lower TCO/chip [E]. ASIC shipments are growing **~44.">40%+ growth holds through the Blackwell ramp — check live figures, as these move weekly.
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.
Key risk to the numbers: estimates have been revised upward almost every quarter for two years; any signal of capex digestion among hyperscalers or China/export restrictions tightening would compress both the EPS estimate and the multiple simultaneously.
Price targets: - Consensus 12-month target: roughly $170–180 (post-10:1 split adjusted) - Range: bears around $110–120, bulls up to $200–220 - Rating skew: overwhelmingly Buy/Overweight (~90% of covering analysts); very few Sells - Implied upside from early-2025 trading levels (~$130–140): ~25–30%
Analysts are bullish with room to run: the average price target is [estimate]">$345.21 — about [state]">50% above today's [state]">$230.36 — and the forward P/E sits at [state]">24.9x, cheap by NVDA's own history.
The numbers (all consensus estimates, not reported figures):
The earnings trajectory behind the multiple: consensus EPS steps up from [estimate]">$9.26 in FY2027 (30 analysts) to [estimate]">$15.64 in FY2028 (32 analysts) to [estimate]">$21.16 in FY2029 (19 analysts), with a small group of 10 analysts at ~[estimate]">$23 for FY2030 [estimate]. Consensus revenue is ~$408B for FY2027, ~$686B for FY2028, ~$904B for FY2029 [estimate]. Next earnings is 2026-11-18, with estimated EPS of [estimate]">$2.47 [estimate].
Revisions are moving the right way. In the latest weekly snapshot, 5 changes were detected and all 5 were upward — FY2028 EPS up [estimate]">0.7%, FY2029 EPS up [estimate]">1.7%, and revenue estimates nudged up 0.3–1.7% across FY2027–FY2029 [estimate]. No downward revisions at all.
How to read the multiple: ~[state]">25x forward on a company that just printed $96.2B of revenue, +106% year-over-year, in Q2 FY2027 [Q2 FY2027] is historically cheap for NVDA — the research view pegs its own five-year average forward P/E around [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 [F]. The market is essentially pricing in deceleration, not disputing the current numbers. One offset worth knowing: a reverse DCF — a model that backs out the growth the current price implies — still requires ~[state]">29.9% annual free-cash-flow growth for a decade [state], so "cheap on P/E" doesn't mean "no expectations embedded." And guidance is for gross margin to trough at 71–72% in Q4 FY2027 on memory costs [guidance], which is the near-term risk to those EPS estimates.
What to watch: (1) the November 18 print against the [estimate]">$2.47 consensus EPS; (2) whether the 71–72% Q4 gross-margin guidance holds or worsens — a slip would pressure FY2028 EPS; (3) the revision trend — five straight upward moves is the current tailwind, and a flip would be an early warning.
Bottom line: analysts see ~[state]">50% upside to an average target of [estimate]">$345 against a [state]">24.9x forward P/E — a bullish consensus built on EPS roughly doubling over the next two fiscal years, with revisions still moving 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.
Analysts are bullish with room to run: the average price target is $345.21 – about 50% above today's $230.36 – and the forward P/E sits at 24.9x, cheap by NVDA's own history.
The numbers (all consensus estimates, not reported figures): - Forward P/E ≈ 24.9x – price of $230.36 divided by estimated FY2027 EPS of $9.26 (30 analysts) . Note the basis: this is a forward multiple on consensus earnings, not trailing; the trailing P/E is 29.1x (TTM through Q2 FY2027, the quarter ended July 2026) . - Price targets: average $345.21, median $322.50, range $270–$515 across 22 analysts . Even the low end of the range sits ~17% above the current price. - Ratings skew hard positive: 2 strong-buy, 58 buy, 16 hold, just 3 sell . - Implied upside to the average target: 49.9% .
The numbers (all consensus estimates, not reported figures): - Forward P/E ≈ 24.9x – price of $230.36 divided by estimated FY2027 EPS of $9.26 (30 analysts) . Note the basis: this is a forward multiple on consensus earnings, not trailing; the trailing P/E is 29.1x (TTM through Q2 FY2027, the quarter ended July 2026) . - Price targets: average $345.21, median $322.50, range $270–$515 across 22 analysts . Even the low end of the range sits ~17% above the current price. - Ratings skew hard positive: 2 strong-buy, 58 buy, 16 hold, just 3 sell . - Implied upside to the average target: 49.9% .
Analysts are bullish with room to run: the average price target is $345.21 – about 50% above today's $230.36 – and the forward P/E sits at 24.9x, cheap by NVDA's own history.
The numbers (all consensus estimates, not reported figures): - Forward P/E ≈ 24.9x – price of $230.36 divided by estimated FY2027 EPS of $9.26 (30 analysts) . Note the basis: this is a forward multiple on consensus earnings, not trailing; the trailing P/E is 29.1x (TTM through Q2 FY2027, the quarter ended July 2026) . - Price targets: average $345.21, median $322.50, range $270–$515 across 22 analysts . Even the low end of the range sits ~17% above the current price. - Ratings skew hard positive: 2 strong-buy, 58 buy, 16 hold, just 3 sell . - Implied upside to the average target: 49.9% .
The earnings trajectory behind the multiple: consensus EPS steps up from $9.26 in FY2027 (30 analysts) to $15.64 in FY2028 (32 analysts) to $21.16 in FY2029 (19 analysts), with a small group of 10 analysts at ~$23 for FY2030 . Consensus revenue is ~$408B for FY2027, ~$686B for FY2028, ~$904B for FY2029 . Next earnings is 2026-11-18, with estimated EPS of $2.47 .
The numbers (all consensus estimates, not reported figures): - Forward P/E ≈ 24.9x – price of $230.36 divided by estimated FY2027 EPS of $9.26 (30 analysts) . Note the basis: this is a forward multiple on consensus earnings, not trailing; the trailing P/E is 29.1x (TTM through Q2 FY2027, the quarter ended July 2026) . - Price targets: average $345.21, median $322.50, range $270–$515 across 22 analysts . Even the low end of the range sits ~17% above the current price. - Ratings skew hard positive: 2 strong-buy, 58 buy, 16 hold, just 3 sell . - Implied upside to the average target: 49.9% .
Revisions are moving the right way. In the latest weekly snapshot, 5 changes were detected and all 5 were upward – FY2028 EPS up 0.7%, FY2029 EPS up 1.7%, and revenue estimates nudged up 0.3–1.7% across FY2027–FY2029 . No downward revisions at all.
The numbers (all consensus estimates, not reported figures): - Forward P/E ≈ 24.9x – price of $230.36 divided by estimated FY2027 EPS of $9.26 (30 analysts) . Note the basis: this is a forward multiple on consensus earnings, not trailing; the trailing P/E is 29.1x (TTM through Q2 FY2027, the quarter ended July 2026) . - Price targets: average $345.21, median $322.50, range $270–$515 across 22 analysts . Even the low end of the range sits ~17% above the current price. - Ratings skew hard positive: 2 strong-buy, 58 buy, 16 hold, just 3 sell . - Implied upside to the average target: 49.9% .
Q3 FY2025 (quarter ended Oct 27, 2024, reported Nov 20) — this is the most recent I can reliably cite; verify if there's been a newer print.
Key qualitative points: Blackwell shipments began ramping, demand described as "staggering" with supply constrained into FY2026; Hopper (H200) still growing in the meantime. Hopper-to-Blackwell margin compression and customer concentration (hyperscalers ~[state]">50% of DC revenue) remain the main watch items.
Bottom line: $35.1B revenue, +94% YoY, another beat-and-raise — growth decelerating off peak but still exceptional, with Blackwell as the FY2026 driver.
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 most recent quarter — Q2 FY2027, ended July 26, 2026 — was a blowout: reported revenue of $96.2B, up 106% year-over-year, with GAAP gross margin of [state]">75.0% and GAAP net income of $59.7B. [Q2 FY2027]
The print versus the bar. Against consensus, reported 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 [F, earnings recap]: non-GAAP EPS $2.22 vs $2.09; revenue $96.2B vs $92.3B; shares +8.7% next">$2.09 by 6.2% and revenue of $96.2B beat $92.3B by 4.3% — extending a streak to 7-for-7 double beats [earnings recap agent]. The scorecard verdict was "beat," not "beat-and-raise," because guided Q3 FY2027 revenue of [guidance]">$108.0B came in 0.6% below next-quarter consensus of $108.65B [earnings recap agent]. Shares still rose [earnings recap agent]">8.7% the next session on [earnings recap agent]">2.6x normal volume — the largest single-day move in the last seven prints [earnings recap agent].
The trend. The tape is re-accelerating, not fading: $44.1B → $46.7B → $57.0B → $81.6B → $96.2B over five quarters [Q1 FY2026][Q2 FY2027]. Gross margin ticked up to [state]">75.0% from [Q1 FY2027]">74.9% in Q1 FY2027 and [Q2 FY2026]">72.4% in the year-ago quarter [Q2 FY2026][Q2 FY2027].
The balance sheet moved more than the P&L. This is what changed under the hood:
The forward frame. Guided Q3 gross margin is [guidance]">74% ±50bp, but management guided a 71–72% trough in Q4 FY2027 and 72–73% for FY2028 on memory-cost inflation — a structural reset, not a one-off [guidance][F09]. FY2028 revenue is guided up ~[earnings recap agent]">70%, explicitly supply-constrained against demand management estimates at ~100% [F07].
What to watch: whether Q3 prints [guidance]">$108B+ and holds [guidance]">74% margin (report due 2026-11-18, estimated EPS [estimate]">$2.47) [estimate]; whether DSO keeps stretching past 60 days; and confirmation of the Q4 margin trough at 71–72%.
Bottom line: an enormous beat-and-hold quarter — $96.2B at [state]">75% margins — but the quarter's real news was [Q2 FY2027]">$25B of new debt, a [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, and management's own admission that margins step down from here.
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 most recent quarter – Q2 FY2027, ended July 26, 2026 – was a blowout: reported revenue of $96.2B, up 106% year-over-year, with GAAP gross margin of 75% and GAAP net income of $59.7B. Q2 FY2027
The most recent quarter – Q2 FY2027, ended July 26, 2026 – was a blowout: reported revenue of $96.2B, up 106% year-over-year, with GAAP gross margin of 75% and GAAP net income of $59.7B. Q2 FY2027
The most recent quarter – Q2 FY2027, ended July 26, 2026 – was a blowout: reported revenue of $96.2B, up 106% year-over-year, with GAAP gross margin of 75% and GAAP net income of $59.7B. Q2 FY2027
The most recent quarter – Q2 FY2027, ended July 26, 2026 – was a blowout: reported revenue of $96.2B, up 106% year-over-year, with GAAP gross margin of 75% and GAAP net income of $59.7B. Q2 FY2027
The print versus the bar. Against consensus, reported non-GAAP EPS of $2.22 beat $2.09 by 6.2% and revenue of $96.2B beat $92.3B by 4.3% – extending a streak to 7-for-7 double beats . The scorecard verdict was "beat," not "beat-and-raise," because guided Q3 FY2027 revenue of $108.0B came in 0.6% below next-quarter consensus of $108.65B . Shares still rose 8.7% the next session on 2.6x normal volume – the largest single-day move in the last seven prints .
The print versus the bar. Against consensus, reported non-GAAP EPS of $2.22 beat $2.09 by 6.2% and revenue of $96.2B beat $92.3B by 4.3% – extending a streak to 7-for-7 double beats . The scorecard verdict was "beat," not "beat-and-raise," because guided Q3 FY2027 revenue of $108.0B came in 0.6% below next-quarter consensus of $108.65B . Shares still rose 8.7% the next session on 2.6x normal volume – the largest single-day move in the last seven prints .
The balance sheet moved more than the P&L. This is what changed under the hood: - Debt jumped from $8.5B to $33.4B after issuing $25.0B of senior unsecured notes in June 2026 Q2 FY2027 . - Receivables ballooned to $63.1B – DSO (days sales outstanding, how long customers take to pay) stretched to ~60 days from ~45 days a quarter earlier, a sign collections are lagging shipments Q1 FY2027Q2 FY2027. - Inventories rose to $31.6B from $25.8B, alongside supply and capacity commitments that more than doubled to $279B from $119B the prior quarter – NVIDIA is locking in supply years out . - On the call, management acknowledged nearly $50B of balance-sheet exposure to pre-profit Frontier AI labs, the heart of the "circular financing" debate .
The balance sheet moved more than the P&L. This is what changed under the hood: - Debt jumped from $8.5B to $33.4B after issuing $25.0B of senior unsecured notes in June 2026 Q2 FY2027 . - Receivables ballooned to $63.1B – DSO (days sales outstanding, how long customers take to pay) stretched to ~60 days from ~45 days a quarter earlier, a sign collections are lagging shipments Q1 FY2027Q2 FY2027. - Inventories rose to $31.6B from $25.8B, alongside supply and capacity commitments that more than doubled to $279B from $119B the prior quarter – NVIDIA is locking in supply years out . - On the call, management acknowledged nearly $50B of balance-sheet exposure to pre-profit Frontier AI labs, the heart of the "circular financing" debate .
The forward frame. Guided Q3 gross margin is 74% ±50bp, but management guided a 71–72% trough in Q4 FY2027 and 72–73% for FY2028 on memory-cost inflation – a structural reset, not a one-off . FY2028 revenue is guided up ~70%, explicitly supply-constrained against demand management estimates at ~100% .
NVDA valuation snapshot (based on my last available data; verify current prices before acting):
The headline multiples
How demanding is it, really?
The risks embedded in the multiple
Bottom line: On forward earnings NVDA is more fairly priced than frothy (~30x for [F] — unresolved. AMD's **Helios/MI455X** (Advancing AI, July 22–23) is a legitimate spec-sheet win on memory (31 TB HBM4, ~50% more) but is priced **~40% above** Vera Rubin ($5.0–5.5M vs $3.5–4.0M/rack [E]) and rides dilutive OpenAI/Meta warrants; Google **Ironwood** claims ~44% lower TCO/chip [E]. ASIC shipments are growing **~44.">40%+ growth), but the valuation demands sustained hyperscaler capex and pricing power — it's a bet on the AI buildout continuing, not on a cheap stock.
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.
How demanding is it, really? - Versus its own history: mid-range. NVDA traded at 80x+ forward earnings in 2023 and has compressed as EPS exploded. - Versus mega-cap peers: trades at a premium to Microsoft/Apple (~30x forward) but with 50%+ revenue growth vs. their low-teens – the PEG ratio is actually below 1 on some estimates, which is rare for a mega-cap. - The valuation is essentially a bet on one thing: that data center capex from hyperscalers (~$250B+/year committed) keeps growing through 2026–27 and that NVDA holds 80%+ accelerator share and 55%+ gross margins against AMD and custom ASICs (Google TPU, Amazon Trainium).
How demanding is it, really? - Versus its own history: mid-range. NVDA traded at 80x+ forward earnings in 2023 and has compressed as EPS exploded. - Versus mega-cap peers: trades at a premium to Microsoft/Apple (~30x forward) but with 50%+ revenue growth vs. their low-teens – the PEG ratio is actually below 1 on some estimates, which is rare for a mega-cap. - The valuation is essentially a bet on one thing: that data center capex from hyperscalers (~$250B+/year committed) keeps growing through 2026–27 and that NVDA holds 80%+ accelerator share and 55%+ gross margins against AMD and custom ASICs (Google TPU, Amazon Trainium).
The price embeds ~[state]">29.9% annual free-cash-flow growth for a full decade [state] — that's the cleanest single answer, and it's demanding but not extreme by NVDA's own history. At [state]">$230.36, the market cap is ~[state]">$5.6 trillion, the trailing P/E (TTM through Q2 FY2027, ended July 2026) is [state]">29.1x, P/S is [state]">18.5x, and EV/EBITDA is [state]">27.9x — all modeled on the same TTM basis [state].
The reverse DCF. This model asks: what growth rate justifies today's price? The answer is roughly 29–30% annual FCF growth for ten years at a ~13.9% discount rate — a genuinely high bar, but one the model flags as just under the [watch]">32% threshold where margin for error gets thin [state]. For calibration, NVDA actually grew revenue +65% in FY2026 and +106% y/y in Q2 FY2027 (reported, GAAP) [Q2 FY2027], so the market is pricing in massive deceleration, not continuation.
The cheap-looking side. Forward P/E is ~[state]">24.9x (price ÷ FY2027 consensus EPS of [estimate]">$9.25743, 30 analysts) [estimate] — far below NVDA's ~[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 and roughly the cheapest forward multiple in a decade. Analysts see ~[state]">49.9% upside to the average target of [estimate]">$345.21 [state]. Earnings estimates keep drifting up: FY2028 consensus revenue rose ~[state]">1% and FY2029 EPS ~[estimate]">1.7% in the latest week [estimate]. The stock is roughly flat YTD while AMD is up over 130% — the market is letting earnings grow into the price rather than re-rating it.
The expensive-looking side. On a simple price-to-sales comparison with chip peers, NVDA's premium implies about [state]">-69.5% downside if it traded at the peer median — a model that deliberately ignores business differences, and one that's already tripped a "[watch]">-65% or wider" watch trigger, meaning the premium is stretching, not compressing [state].
What the price is really saying. The puzzle resolves to one of two beliefs: either FY2027 is peak-cycle earnings, or growth persists but shouldn't be capitalized at a normal multiple because of the financing circularity downstream. The tension is visible: insiders have been net sellers (~4.7M shares over six months, zero open-market buys) [state], the consensus target sits ~[state]">50% above the price [estimate], and the guided gross-margin trough of 71–72% in Q4 FY2027 (from [state]">75.0% printed in Q2) is a known compression the model has to absorb.
What to watch: (1) reported FCF each quarter against that ~[state]">30% decade-long requirement; (2) whether the peer P/S gap widens past its current [state]">-69.5%; (3) the Nov-18 earnings print (consensus EPS [estimate]">$2.47) and whether Q4's margin trough confirms at 71–72%.
Bottom line: demanding on absolute cash-flow terms (~[state]">30% FCF growth for a decade is a real ask), but historically cheap on forward earnings — the valuation only breaks if the AI buildout's financing, not demand, cracks.
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 price embeds ~29.9% annual free-cash-flow growth for a full decade – that's the cleanest single answer, and it's demanding but not extreme by NVDA's own history. At $230.36, the market cap is ~$5.6 trillion, the trailing P/E (TTM through Q2 FY2027, ended July 2026) is 29.1x, P/S is 18.5x, and EV/EBITDA is 27.9x – all modeled on the same TTM basis .
The cheap-looking side. Forward P/E is ~24.9x (price ÷ FY2027 consensus EPS of $9.25743, 30 analysts) – far below NVDA's ~63x five-year average and roughly the cheapest forward multiple in a decade. Analysts see ~49.9% upside to the average target of $345.21 . Earnings estimates keep drifting up: FY2028 consensus revenue rose ~1% and FY2029 EPS ~1.7% in the latest week . The stock is roughly flat YTD while AMD is up over 130% – the market is letting earnings grow into the price rather than re-rating it.
The reverse DCF. This model asks: what growth rate justifies today's price? The answer is roughly 29–30% annual FCF growth for ten years at a ~13.9% discount rate – a genuinely high bar, but one the model flags as just under the 32% threshold where margin for error gets thin . For calibration, NVDA actually grew revenue +65% in FY2026 and +106% y/y in Q2 FY2027 (reported, GAAP) Q2 FY2027, so the market is pricing in massive deceleration, not continuation.
The expensive-looking side. On a simple price-to-sales comparison with chip peers, NVDA's premium implies about -69.5% downside if it traded at the peer median – a model that deliberately ignores business differences, and one that's already tripped a "-65% or wider" watch trigger, meaning the premium is stretching, not compressing .
The price embeds ~29.9% annual free-cash-flow growth for a full decade – that's the cleanest single answer, and it's demanding but not extreme by NVDA's own history. At $230.36, the market cap is ~$5.6 trillion, the trailing P/E (TTM through Q2 FY2027, ended July 2026) is 29.1x, P/S is 18.5x, and EV/EBITDA is 27.9x – all modeled on the same TTM basis .
The price embeds ~29.9% annual free-cash-flow growth for a full decade – that's the cleanest single answer, and it's demanding but not extreme by NVDA's own history. At $230.36, the market cap is ~$5.6 trillion, the trailing P/E (TTM through Q2 FY2027, ended July 2026) is 29.1x, P/S is 18.5x, and EV/EBITDA is 27.9x – all modeled on the same TTM basis .
The price embeds ~29.9% annual free-cash-flow growth for a full decade – that's the cleanest single answer, and it's demanding but not extreme by NVDA's own history. At $230.36, the market cap is ~$5.6 trillion, the trailing P/E (TTM through Q2 FY2027, ended July 2026) is 29.1x, P/S is 18.5x, and EV/EBITDA is 27.9x – all modeled on the same TTM basis .
The cheap-looking side. Forward P/E is ~24.9x (price ÷ FY2027 consensus EPS of $9.25743, 30 analysts) – far below NVDA's ~63x five-year average and roughly the cheapest forward multiple in a decade. Analysts see ~49.9% upside to the average target of $345.21 . Earnings estimates keep drifting up: FY2028 consensus revenue rose ~1% and FY2029 EPS ~1.7% in the latest week . The stock is roughly flat YTD while AMD is up over 130% – the market is letting earnings grow into the price rather than re-rating it.
1. Customer concentration & in-house silicon Roughly 40–50% of data center revenue comes from a handful of hyperscalers (Microsoft, Meta, Amazon, Google). All four are deploying custom accelerators (TPU, Trainium, MTIA, Maia). Even partial workload migration off Nvidia GPUs compresses both volume and pricing power.
2. AI capex digestion risk The bull case assumes hyperscaler capex stays on a steep upward trajectory. If AI monetization disappoints and customers slow orders—even for a quarter or two—NVDA's growth multiple would re-rate fast. Order visibility beyond a few quarters is limited, and double-ordering in past cycles has inflated apparent demand.
3. Export controls on China Successive U.S. restrictions have already forced product redesigns (H20), and further tightening remains a live policy risk. China was historically ~20%+ of data center revenue; the effective ceiling on what can be shipped there could drop to near zero under stricter rules.
4. Competition catching up AMD's MI300/MI400 ramp and custom ASICs from Broadcom/Marvell (designed with hyperscalers) are eroding share at the margin. NVLink/CUDA moats remain strong, but inference workloads are more portable than training—where much future demand growth sits.
5. Supply chain & margin dependency Concentration on TSMC CoWoS packaging and HBM supply (SK Hynix, Micron, Samsung) creates bottlenecks and cost pressure. Gross margins in the mid-70s% assume continued premium pricing; competition plus rising HBM content could compress them toward the low 70s or below.
6. Valuation & sentiment risk Trading at a premium multiple on peak-cycle earnings, the stock is highly sensitive to any guidance miss. It's also a crowded institutional and index-heavy position, amplifying downside in a risk-off tape.
7. Transition execution Rubin (2026) ramp execution and networking integration matter; any delay or thermal/yield issue on Blackwell Ultra→Rubin transitions would hit near-term numbers.
Bottom line: The core risk isn't competition killing the moat—it's hyperscaler capex stalling or shifting in-house while NVDA trades on flawless execution, making guidance cadence the single most important near-term variable.
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. Export controls on China Successive U.S. restrictions have already forced product redesigns (H20), and further tightening remains a live policy risk. China was historically ~20%+ of data center revenue; the effective ceiling on what can be shipped there could drop to near zero under stricter rules.
1. Customer concentration & in-house silicon Roughly 40–50% of data center revenue comes from a handful of hyperscalers (Microsoft, Meta, Amazon, Google). All four are deploying custom accelerators (TPU, Trainium, MTIA, Maia). Even partial workload migration off Nvidia GPUs compresses both volume and pricing power.
5. Supply chain & margin dependency Concentration on TSMC CoWoS packaging and HBM supply (SK Hynix, Micron, Samsung) creates bottlenecks and cost pressure. Gross margins in the mid-70s% assume continued premium pricing; competition plus rising HBM content could compress them toward the low 70s or below.
5. Supply chain & margin dependency Concentration on TSMC CoWoS packaging and HBM supply (SK Hynix, Micron, Samsung) creates bottlenecks and cost pressure. Gross margins in the mid-70s% assume continued premium pricing; competition plus rising HBM content could compress them toward the low 70s or below.
4. Competition catching up AMD's MI300/MI400 ramp and custom ASICs from Broadcom/Marvell (designed with hyperscalers) are eroding share at the margin. NVLink/CUDA moats remain strong, but inference workloads are more portable than training–where much future demand growth sits.
2. AI capex digestion risk The bull case assumes hyperscaler capex stays on a steep upward trajectory. If AI monetization disappoints and customers slow orders–even for a quarter or two–NVDA's growth multiple would re-rate fast. Order visibility beyond a few quarters is limited, and double-ordering in past cycles has inflated apparent demand.
NVDA's risk picture has shifted: the "is demand real?" debate is over ($96.2B printed in Q2 FY2027, ended July 2026 [Q2 FY2027]), and the live risks are now about financing, supply, and concentration — the things that could break the model from underneath. Here are the biggest, ranked by how material they look today.
1. Financing risk — NVIDIA is now funding its own ecosystem. This is the newest and largest change. The latest 10-Q (filed 2026-08-26) discloses guarantees capped at $105 billion to SB Energy Corp. for a land, power, and shell buildout, out of [10-Q notes]">$108.5B total maximum guarantee exposure [10-Q notes]. Separately, NVIDIA signed MOUs to mobilize $500 billion of third-party AI infrastructure capital, with optional residual-value support on some projects — these "may not lead to definitive agreements," so treat the headline as aspirational [SEC filings]. Add the new AI-cloud partner model: NVIDIA commits to buy unsold capacity from AI cloud partners, roughly [Q3 FY2026]">$36 billion of six-year commitments [10-Q notes], plus a reported ~$250B Nvidia-OpenAI deal that raised "circular financing" concerns and drove a ~4% stock drop in July [watch]. The through-line: if third-party AI demand underperforms, NVIDIA is now contractually on the hook in ways it wasn't two quarters ago. The canary here is CoreWeave credit default swaps — insurance costs on a key "neocloud" customer — which peaked around [watch]">881bps in Dec-2025 and sit near [watch]">452bps; a move back above [watch]">600bps would signal renewed stress [watch].
2. Supply-chain and inventory risk — massive commitments against tight components. Supply and capacity commitments more than doubled in one quarter, 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 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 [10-Q notes]. NVIDIA is fabless — it designs chips but outsources manufacturing — so it depends on third-party foundries and has no direct control over quantity, yields, or delivery [risk]. Inventories are already climbing: [Q1 FY2026]">$11.3B in Q1 FY2026 → [Q2 FY2027]">$31.6B in Q2 FY2027 [Q1 FY2026][Q2 FY2027], and live reporting shows memory suppliers' inventories below 10 days, signaling tight HBM supply [risk]. Management itself guided gross margin down to 71–72% in Q4 FY2027 largely on memory costs, versus [state]">75.0% printed in Q2 — a slip below 73% is the tripwire I'm tracking [watch]. Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" acknowledged, so a product transition is running into a parts shortage [SEC filings].
3. Customer concentration and capex durability. One direct customer was [guidance]">16% of Q2 FY2027 revenue; three were [guidance]">16%, 15%, and [10-Q notes]">13% of first-half revenue [10-Q notes]. Five customers hold [10-Q notes]">22%, [10-Q notes]">14%, [10-Q notes]">13%, [state]">11%, and [10-Q notes]">10% of the receivables balance — which, alongside receivables jumping to [Q2 FY2027]">$63.1B (DSO up to ~60 days from ~45 days in Q1 FY2027), means more of the P&L is riding on fewer counterparties [10-Q notes][Q1 FY2027][Q2 FY2027]. The single highest-signal bear trigger on my list: a coordinated 2027 capex-guidance cut from Microsoft, Amazon, Meta, and Google would flip the thesis outright [watch].
4. Geopolitical / export-control risk — China is now effectively zero, and worsening. The new 10-Q discloses that PRC restrictions prevented H20 sales despite U.S. government licenses, reversing what was expected to be a licensed China revenue stream [SEC filings]. Management's Q3 guidance of [guidance]">$108B ±[guidance]">2% explicitly excludes any China Data Center compute revenue [guidance]. New disclosures add that foreign governments may direct customers to competitors in response to export controls, and that export rules are triggering foreign investigations [SEC filings]. Live coverage underscores this: Huawei is pushing China's domestic chip self-sufficiency drive [risk]. FY2026 already absorbed a ~$4.5B China-related write-down, so this risk has a track record of costing real money.
5. Antitrust and leverage. Regulators worldwide are investigating GPU/AI market competition with burdensome information requests [risk], and the prior quarter's specific China antitrust finding over Mellanox acquisition terms was replaced by generalized export-control language — the specific allegation is no longer disclosed, which reads as de-escalation but not resolution [SEC filings]. On leverage: NVIDIA issued [Q2 FY2027]">$25B of senior notes in June 2026, taking total debt from $8.5B to $33.4B, and added a standalone indebtedness/covenants risk factor for the first time [10-Q notes][SEC filings]. Given ~$74B of H1 operating cash flow, this isn't a solvency issue — it's a flexibility signal, and worth watching alongside the guarantee stack.
Also disclosed, not material to this question: a $2.9B financing-section payment to Groq, Inc.; [10-Q notes]">$51.2B of non-marketable securities; $1.6B of H1 inventory provisions; and zero accrued legal contingencies [10-Q notes].
What to watch: CoreWeave CDS above [watch]">600bps; hyperscaler 2027 capex guides in upcoming earnings season; Q4 FY2027 gross margin vs the guided 71–72% trough [watch].
Bottom line: demand isn't the risk anymore — the risk is that NVIDIA's own balance sheet (guarantees, capacity purchase 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 of supply obligations) is now underwriting the buildout, so a demand wobble hits harder and faster than it would have a year ago.
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.
4. Geopolitical / export-control risk – China is now effectively zero, and worsening. The new 10-Q discloses that PRC restrictions prevented H20 sales despite U.S. government licenses, reversing what was expected to be a licensed China revenue stream . Management's Q3 guidance of $108B ±2% explicitly excludes any China Data Center compute revenue . New disclosures add that foreign governments may direct customers to competitors in response to export controls, and that export rules are triggering foreign investigations . Live coverage underscores this: Huawei is pushing China's domestic chip self-sufficiency drive . FY2026 already absorbed a ~$4.5B China-related write-down, so this risk has a track record of costing real money.
3. Customer concentration and capex durability. One direct customer was 16% of Q2 FY2027 revenue; three were 16%, 15%, and 13% of first-half revenue . Five customers hold 22%, 14%, 13%, 11%, and 10% of the receivables balance – which, alongside receivables jumping to $63.1B (DSO up to ~60 days from ~45 days in Q1 FY2027), means more of the P&L is riding on fewer counterparties Q1 FY2027Q2 FY2027. The single highest-signal bear trigger on my list: a coordinated 2027 capex-guidance cut from Microsoft, Amazon, Meta, and Google would flip the thesis outright .
2. Supply-chain and inventory risk – massive commitments against tight components. Supply and capacity commitments more than doubled in one quarter, from $119B to $279 billion . NVIDIA is fabless – it designs chips but outsources manufacturing – so it depends on third-party foundries and has no direct control over quantity, yields, or delivery . Inventories are already climbing: $11.3B in Q1 FY2026 → $31.6B in Q2 FY2027 Q1 FY2026Q2 FY2027, and live reporting shows memory suppliers' inventories below 10 days, signaling tight HBM supply . Management itself guided gross margin down to 71–72% in Q4 FY2027 largely on memory costs, versus 75% printed in Q2 – a slip below 73% is the tripwire I'm tracking . Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" acknowledged, so a product transition is running into a parts shortage .
1. Financing risk – NVIDIA is now funding its own ecosystem. This is the newest and largest change. The latest 10-Q (filed 2026-08-26) discloses guarantees capped at $105 billion to SB Energy Corp. for a land, power, and shell buildout, out of $108.5B total maximum guarantee exposure . Separately, NVIDIA signed MOUs to mobilize $500 billion of third-party AI infrastructure capital, with optional residual-value support on some projects – these "may not lead to definitive agreements," so treat the headline as aspirational . Add the new AI-cloud partner model: NVIDIA commits to buy unsold capacity from AI cloud partners, roughly $36 billion of six-year commitments , plus a reported ~$250B Nvidia-OpenAI deal that raised "circular financing" concerns and drove a ~4% stock drop in July . The through-line: if third-party AI demand underperforms, NVIDIA is now contractually on the hook in ways it wasn't two quarters ago. The canary here is CoreWeave credit default swaps – insurance costs on a key "neocloud" customer – which peaked around 881bps in Dec-2025 and sit near 452bps; a move back above 600bps would signal renewed stress .
2. Supply-chain and inventory risk – massive commitments against tight components. Supply and capacity commitments more than doubled in one quarter, from $119B to $279 billion . NVIDIA is fabless – it designs chips but outsources manufacturing – so it depends on third-party foundries and has no direct control over quantity, yields, or delivery . Inventories are already climbing: $11.3B in Q1 FY2026 → $31.6B in Q2 FY2027 Q1 FY2026Q2 FY2027, and live reporting shows memory suppliers' inventories below 10 days, signaling tight HBM supply . Management itself guided gross margin down to 71–72% in Q4 FY2027 largely on memory costs, versus 75% printed in Q2 – a slip below 73% is the tripwire I'm tracking . Vera Rubin began production shipments in Q3 FY2027 with "certain supply constraints" acknowledged, so a product transition is running into a parts shortage .
4. Geopolitical / export-control risk – China is now effectively zero, and worsening. The new 10-Q discloses that PRC restrictions prevented H20 sales despite U.S. government licenses, reversing what was expected to be a licensed China revenue stream . Management's Q3 guidance of $108B ±2% explicitly excludes any China Data Center compute revenue . New disclosures add that foreign governments may direct customers to competitors in response to export controls, and that export rules are triggering foreign investigations . Live coverage underscores this: Huawei is pushing China's domestic chip self-sufficiency drive . FY2026 already absorbed a ~$4.5B China-related write-down, so this risk has a track record of costing real money.
3. Customer concentration and capex durability. One direct customer was 16% of Q2 FY2027 revenue; three were 16%, 15%, and 13% of first-half revenue . Five customers hold 22%, 14%, 13%, 11%, and 10% of the receivables balance – which, alongside receivables jumping to $63.1B (DSO up to ~60 days from ~45 days in Q1 FY2027), means more of the P&L is riding on fewer counterparties Q1 FY2027Q2 FY2027. The single highest-signal bear trigger on my list: a coordinated 2027 capex-guidance cut from Microsoft, Amazon, Meta, and Google would flip the thesis outright .
5. Antitrust and leverage. Regulators worldwide are investigating GPU/AI market competition with burdensome information requests , and the prior quarter's specific China antitrust finding over Mellanox acquisition terms was replaced by generalized export-control language – the specific allegation is no longer disclosed, which reads as de-escalation but not resolution . On leverage: NVIDIA issued $25B of senior notes in June 2026, taking total debt from $8.5B to $33.4B, and added a standalone indebtedness/covenants risk factor for the first time . Given ~$74B of H1 operating cash flow, this isn't a solvency issue – it's a flexibility signal, and worth watching alongside the guarantee stack.
5. Antitrust and leverage. Regulators worldwide are investigating GPU/AI market competition with burdensome information requests , and the prior quarter's specific China antitrust finding over Mellanox acquisition terms was replaced by generalized export-control language – the specific allegation is no longer disclosed, which reads as de-escalation but not resolution . On leverage: NVIDIA issued $25B of senior notes in June 2026, taking total debt from $8.5B to $33.4B, and added a standalone indebtedness/covenants risk factor for the first time . Given ~$74B of H1 operating cash flow, this isn't a solvency issue – it's a flexibility signal, and worth watching alongside the guarantee stack.
NVDA guidance tracking — delivered, revised, missed (last ~6 quarters through Q3 FY2026, reported Nov 2025)
Delivered (the dominant pattern)
Revised
Missed / watch items
Scorecard: On the metrics management controls — revenue, margin, product ramp — the record is near-flawless. The only material revisions have been externally forced (China export policy).
Bottom line: NVDA has beaten its own guidance every quarter for over two years and delivered on Blackwell and margin promises; the only meaningful walk-backs are China-related and policy-driven, not execution-driven.
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.
Delivered (the dominant pattern) - Revenue guidance: beaten every quarter. NVDA has topped its own revenue guide for 10+ consecutive quarters. Examples: guided ~$37.5B for Q4 FY25, delivered ~$39.3B; guided ~$43B for Q1 FY26, delivered ~$44.1B; guided ~$45B for Q2 FY26, delivered ~$46.7B; guided $54B for Q3 FY26, delivered $57.0B. Q4 FY26 guide: ~$65B – a raise well above the ~$62B consensus. - Blackwell ramp: Promised "several billion" in Blackwell revenue in Q4 FY25 – delivered ~$11B. Promised production shipments of Blackwell Ultra (GB300) in H2 FY26 – on track per Q2/Q3 calls. - Gross margin recovery: After the Blackwell transition dip (GM fell to ~61% in Q4 FY25), management guided back to "mid-70s" by year-end – delivered ~73.4% non-GAAP in Q3 FY26. - Networking: Promised strong growth; delivered ~$8.2B in Q3 FY26, +162% YoY.
Delivered (the dominant pattern) - Revenue guidance: beaten every quarter. NVDA has topped its own revenue guide for 10+ consecutive quarters. Examples: guided ~$37.5B for Q4 FY25, delivered ~$39.3B; guided ~$43B for Q1 FY26, delivered ~$44.1B; guided ~$45B for Q2 FY26, delivered ~$46.7B; guided $54B for Q3 FY26, delivered $57.0B. Q4 FY26 guide: ~$65B – a raise well above the ~$62B consensus. - Blackwell ramp: Promised "several billion" in Blackwell revenue in Q4 FY25 – delivered ~$11B. Promised production shipments of Blackwell Ultra (GB300) in H2 FY26 – on track per Q2/Q3 calls. - Gross margin recovery: After the Blackwell transition dip (GM fell to ~61% in Q4 FY25), management guided back to "mid-70s" by year-end – delivered ~73.4% non-GAAP in Q3 FY26. - Networking: Promised strong growth; delivered ~$8.2B in Q3 FY26, +162% YoY.
NVDA guidance tracking – delivered, revised, missed (last ~6 quarters through Q3 FY2026, reported Nov 2025)
NVDA management's recent commitment record is a study in two halves: on products, capital return and sovereign AI they've over-delivered; on operating-expense discipline they've revised guidance upward three separate times. Across 6 earnings calls (May 2025 to August 2026), 31 commitments were tracked — 17 resolved, 14 still pending [transcripts].
Delivered (and mostly beaten):
Revised upward (withdrawn targets) — all on OpEx:
No outright misses appear in the record — the pattern is delivery on revenue/product promises paired with serial upward OpEx revisions. Worth noting the flip side: each OpEx raise accompanied revenue guidance that was also rising, so the revisions reflect investment scaling, not cost overruns on flat demand.
What to watch: the 14 pending commitments, and whether the Q4 FY2027 gross-margin trough lands inside the guided 71–72% range [watch] — that would be the first margin test of the memory-inflation era. Also the next print on 2026-11-18 (estimated EPS [estimate]">$2.47, consensus) for any further OpEx creep [estimate].
Bottom line: management's credibility on demand and product execution is intact — every revenue-side promise delivered or beaten — but they've moved the OpEx goalposts three times in five quarters, so treat the "low 50s" FY2027 OpEx growth line as a floor, not a ceiling.
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.
- FY2026 OpEx growth: guided "mid-thirties" (Feb and May 2025 calls), then raised to "high thirties" in August 2025 . - FY2027 OpEx growth: guided "low 40s" (Feb 2026), raised to "upper forties" (May 2026), raised again to "low 50s" (Aug 2026) – management citing a broadening product portfolio and heavier internal AI-tool usage .
- FY2026 OpEx growth: guided "mid-thirties" (Feb and May 2025 calls), then raised to "high thirties" in August 2025 . - FY2027 OpEx growth: guided "low 40s" (Feb 2026), raised to "upper forties" (May 2026), raised again to "low 50s" (Aug 2026) – management citing a broadening product portfolio and heavier internal AI-tool usage .
- Vera Rubin production shipments – promised second-half 2026 start; delivered in August 2026, early in the window, with purchase orders already in hand from every major hyperscaler, AI cloud and system OEM . That's the key product-ramp promise kept. - Mid-seventies gross margin exit for FY2026 – promised August 2025; delivered, with Q4 FY2026 non-GAAP gross margin of 75.2% . - Sovereign AI >$20B for FY2026 – came in at over $30B, more than triple the prior year and 50% above the target . - Capital return – a standing commitment, delivered at record scale: $26B returned in Q2 FY2027 alone ($20B buybacks, $6B dividends), with $39.0B of buybacks fiscal year-to-date and $99.3B of repurchase authorization remaining Q2 FY2027 .
- Vera Rubin production shipments – promised second-half 2026 start; delivered in August 2026, early in the window, with purchase orders already in hand from every major hyperscaler, AI cloud and system OEM . That's the key product-ramp promise kept. - Mid-seventies gross margin exit for FY2026 – promised August 2025; delivered, with Q4 FY2026 non-GAAP gross margin of 75.2% . - Sovereign AI >$20B for FY2026 – came in at over $30B, more than triple the prior year and 50% above the target . - Capital return – a standing commitment, delivered at record scale: $26B returned in Q2 FY2027 alone ($20B buybacks, $6B dividends), with $39.0B of buybacks fiscal year-to-date and $99.3B of repurchase authorization remaining Q2 FY2027 .
NVDA management's recent commitment record is a study in two halves: on products, capital return and sovereign AI they've over-delivered; on operating-expense discipline they've revised guidance upward three separate times. Across 6 earnings calls (May 2025 to August 2026), 31 commitments were tracked – 17 resolved, 14 still pending .
NVDA management's recent commitment record is a study in two halves: on products, capital return and sovereign AI they've over-delivered; on operating-expense discipline they've revised guidance upward three separate times. Across 6 earnings calls (May 2025 to August 2026), 31 commitments were tracked – 17 resolved, 14 still pending .
Caveat: My data has a cutoff, so "latest" here means the 10-Q for Q2 FY2026 (quarter ended July 27, 2025) versus the Q1 FY2026 filing (quarter ended April 27, 2025). If a Q3 FY26 10-Q has since been filed, treat this as the framework for comparing it.
China / export controls — the biggest delta.
Margins inflected back up.
Revenue mix.
Customer concentration got more explicit.
Bottom line: The Q2 10-Q removed the Q1 China overhang, proved margins recover to ~72–73% without H20, and the new disclosures to watch are rising purchase commitments (demand signal) and customer concentration (thesis risk) — the China license remains unresolved upside.
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.
Customer concentration got more explicit. - Two direct customers represented roughly 23% and 16% of Q2 revenue (Customer A and B), and indirect/end-customer concentration (cloud hyperscalers) is disclosed at similar scale. Concentration risk is rising, not falling.
Eleven material disclosures changed in the 10-Q filed 2026-08-26 (quarter ended 2026-07-26) versus the prior 10-Q filed 2026-05-20, and four of them genuinely matter for the thesis: the financing buildout, the supply-commitment surge, the new AI-cloud capacity guarantees, and the China reversal. [SEC filings]
1. The financing architecture (HIGH, new) — NVIDIA is now underwriting its own demand ecosystem. Two linked disclosures show this. First, MOUs with large capital providers "to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital" for AI infrastructure — with NVIDIA optionally providing "limited residual-value support" on specific projects. These are preliminary arrangements and may not become definitive. [SEC filings] Second, a new business model with select AI cloud partners where NVIDIA commits to buy unsold capacity: six-year commitments totaling [Q3 FY2026]">$36 billion as of 2026-07-26, shrinking as third parties use the capacity. [SEC filings] The matching risk factor makes the obligation explicit: NVIDIA guarantees capacity purchases if unsold to third parties. [SEC filings] This ties directly to the "circular financing" worry — NVIDIA capital funding its own demand — already a live watch item after the reported ~$250B OpenAI financing/compute deal drove a ~4% one-day drop on 2026-07-27. [watch] The 10-Q doesn't confirm or deny the OpenAI deal specifically; there's no exact revenue-at-risk figure from these guarantees, but the direction is clear: contingent obligations are stacking on top of a reported [10-Q notes]">$108.5B total maximum guarantee exposure, including $105B to SB Energy Corp. for land, power and shell buildout. [10-Q notes]
2. Supply commitments more than doubled (HIGH). Supply and capacity commitments jumped 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 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]">$279 billion as of July 26, 2026 — part of [10-Q notes]">$366 billion in total future purchase and other commitments, with $220 billion due within the rest of this fiscal year and year one. [10-Q notes] Inventory is already building: [Q2 FY2027]">$31.6B in Q2 FY2027 (quarter ended July 2026), up from [Q1 FY2026]">$11.3B in the year-ago quarter, with reported inventory write-downs of $784M in the quarter and $1.6B for the first half in Cost of revenue. [Q2 FY2027][10-Q notes] NVIDIA is fabless, so this is a purchase-obligation and inventory-risk story, not an input-cost-to-COGS story — if demand normalizes, 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 locked-in supply becomes the swing liability.
3. China: the reversal (HIGH). Two risk-factor changes: PRC government restrictions prevented H20 sales despite U.S. government licenses, reversing the prior expectation of a licensed China revenue stream; and the specific China antitrust preliminary finding (alleged violation of Mellanox acquisition commitments) was replaced with generalized export-control language. [SEC filings] Management's Q3 FY2027 guidance already excludes any Data Center compute revenue from China — guided, not reported — so the base case is consistent, but the optionality is now gone rather than deferred. [guidance] New risk factors add that export rules may trigger foreign government inquiries and that overseas governments may direct customers to competitors. [SEC filings]
4. Vera Rubin shipped, supply constrained (HIGH). Vera Rubin began production shipments in Q3 FY2027, with NVIDIA shipping both Blackwell and Rubin and "currently experiencing certain supply constraints." [SEC filings] That matters against reported Q2 FY2027 Data Center revenue of $89.0 billion, up 117% year-over-year and [guidance]">18% sequentially on Blackwell Ultra. [SEC filings] And the guided gross margin reset to 71–72% in Q4 FY2027 (guided, from the 2026-08-26 call) — versus the reported [state]">75.0% in Q2 FY2027 — is attributed to memory cost inflation during this ramp; that's the margin trigger now scheduled to fire. [watch][Q2 FY2027]
Context that frames all of it: receivables more than doubled to [Q2 FY2027]">$63.1B with DSO stretching to ~60 days from ~54 a year ago; debt quadrupled to [Q3 FY2026]">$33.4B after a [Q2 FY2027]">$25B seven-tranche senior unsecured notes issue in June 2026; non-marketable securities more than doubled to [10-Q notes]">$51.2B with $12.5B of H1 unrealized gains on public equities; deferred revenue additions jumped to [10-Q notes]">$17.7B including [10-Q notes]">$15.6B of customer advances; and customer concentration eased slightly — one direct customer at [guidance]">16% of Q2 revenue versus three at 21%/17%/[guidance]">16% in Q1. [Q2 FY2027][10-Q notes][SEC filings] A new standalone indebtedness/covenants risk factor accompanies the debt build. [SEC filings]
For the thesis: the filing strengthens the demand story (customer advances, Rubin ramp, $500B capital mobilization) while materially raising the contingent-liability and execution stakes — exactly the near-flawless execution the modeled reverse-DCF [state]">29.9% implied FCF growth requires. [reverse-DCF]
Also disclosed, not material to this question: a legal loss-contingency accrual of [10-Q notes]">$0 (liabilities "reasonably possible" but "not probable"), $99.3B of remaining buyback authorization, and a shift in the marketable-securities portfolio to U.S. government-only debt holdings. [10-Q notes][SEC filings]
Bottom line: the 10-Q's biggest changes are NVIDIA formally financing its own demand (capacity guarantees, $500B capital MOUs) and locking in [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 just as China optionality died and margins reset lower — the bull case got bigger, but so did the cost of being wrong.
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. The financing architecture (HIGH, new) – NVIDIA is now underwriting its own demand ecosystem. Two linked disclosures show this. First, MOUs with large capital providers "to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital" for AI infrastructure – with NVIDIA optionally providing "limited residual-value support" on specific projects. These are preliminary arrangements and may not become definitive. Second, a new business model with select AI cloud partners where NVIDIA commits to buy unsold capacity: six-year commitments totaling $36 billion as of 2026-07-26, shrinking as third parties use the capacity. The matching risk factor makes the obligation explicit: NVIDIA guarantees capacity purchases if unsold to third parties. This ties directly to the "circular financing" worry – NVIDIA capital funding its own demand – already a live watch item after the reported ~$250B OpenAI financing/compute deal drove a ~4% one-day drop on 2026-07-27. The 10-Q doesn't confirm or deny the OpenAI deal specifically; there's no exact revenue-at-risk figure from these guarantees, but the direction is clear: contingent obligations are stacking on top of a reported $108.5B total maximum guarantee exposure, including $105B to SB Energy Corp. for land, power and shell buildout.
2. Supply commitments more than doubled (HIGH). Supply and capacity commitments jumped from $119 billion last quarter to $279 billion as of July 26, 2026 – part of $366 billion in total future purchase and other commitments, with $220 billion due within the rest of this fiscal year and year one. Inventory is already building: $31.6B in Q2 FY2027 (quarter ended July 2026), up from $11.3B in the year-ago quarter, with reported inventory write-downs of $784M in the quarter and $1.6B for the first half in Cost of revenue. Q2 FY2027 NVIDIA is fabless, so this is a purchase-obligation and inventory-risk story, not an input-cost-to-COGS story – if demand normalizes, the $279B of locked-in supply becomes the swing liability.
1. The financing architecture (HIGH, new) – NVIDIA is now underwriting its own demand ecosystem. Two linked disclosures show this. First, MOUs with large capital providers "to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital" for AI infrastructure – with NVIDIA optionally providing "limited residual-value support" on specific projects. These are preliminary arrangements and may not become definitive. Second, a new business model with select AI cloud partners where NVIDIA commits to buy unsold capacity: six-year commitments totaling $36 billion as of 2026-07-26, shrinking as third parties use the capacity. The matching risk factor makes the obligation explicit: NVIDIA guarantees capacity purchases if unsold to third parties. This ties directly to the "circular financing" worry – NVIDIA capital funding its own demand – already a live watch item after the reported ~$250B OpenAI financing/compute deal drove a ~4% one-day drop on 2026-07-27. The 10-Q doesn't confirm or deny the OpenAI deal specifically; there's no exact revenue-at-risk figure from these guarantees, but the direction is clear: contingent obligations are stacking on top of a reported $108.5B total maximum guarantee exposure, including $105B to SB Energy Corp. for land, power and shell buildout.
Context that frames all of it: receivables more than doubled to $63.1B with DSO stretching to ~60 days from ~54 a year ago; debt quadrupled to $33.4B after a $25B seven-tranche senior unsecured notes issue in June 2026; non-marketable securities more than doubled to $51.2B with $12.5B of H1 unrealized gains on public equities; deferred revenue additions jumped to $17.7B including $15.6B of customer advances; and customer concentration eased slightly – one direct customer at 16% of Q2 revenue versus three at 21%/17%/16% in Q1. Q2 FY2027 A new standalone indebtedness/covenants risk factor accompanies the debt build.
3. China: the reversal (HIGH). Two risk-factor changes: PRC government restrictions prevented H20 sales despite U.S. government licenses, reversing the prior expectation of a licensed China revenue stream; and the specific China antitrust preliminary finding (alleged violation of Mellanox acquisition commitments) was replaced with generalized export-control language. Management's Q3 FY2027 guidance already excludes any Data Center compute revenue from China – guided, not reported – so the base case is consistent, but the optionality is now gone rather than deferred. New risk factors add that export rules may trigger foreign government inquiries and that overseas governments may direct customers to competitors.
3. China: the reversal (HIGH). Two risk-factor changes: PRC government restrictions prevented H20 sales despite U.S. government licenses, reversing the prior expectation of a licensed China revenue stream; and the specific China antitrust preliminary finding (alleged violation of Mellanox acquisition commitments) was replaced with generalized export-control language. Management's Q3 FY2027 guidance already excludes any Data Center compute revenue from China – guided, not reported – so the base case is consistent, but the optionality is now gone rather than deferred. New risk factors add that export rules may trigger foreign government inquiries and that overseas governments may direct customers to competitors.
4. Vera Rubin shipped, supply constrained (HIGH). Vera Rubin began production shipments in Q3 FY2027, with NVIDIA shipping both Blackwell and Rubin and "currently experiencing certain supply constraints." That matters against reported Q2 FY2027 Data Center revenue of $89.0 billion, up 117% year-over-year and 18% sequentially on Blackwell Ultra. And the guided gross margin reset to 71–72% in Q4 FY2027 (guided, from the 2026-08-26 call) – versus the reported 75% in Q2 FY2027 – is attributed to memory cost inflation during this ramp; that's the margin trigger now scheduled to fire. Q2 FY2027
Context that frames all of it: receivables more than doubled to $63.1B with DSO stretching to ~60 days from ~54 a year ago; debt quadrupled to $33.4B after a $25B seven-tranche senior unsecured notes issue in June 2026; non-marketable securities more than doubled to $51.2B with $12.5B of H1 unrealized gains on public equities; deferred revenue additions jumped to $17.7B including $15.6B of customer advances; and customer concentration eased slightly – one direct customer at 16% of Q2 revenue versus three at 21%/17%/16% in Q1. Q2 FY2027 A new standalone indebtedness/covenants risk factor accompanies the debt build.
1. The financing architecture (HIGH, new) – NVIDIA is now underwriting its own demand ecosystem. Two linked disclosures show this. First, MOUs with large capital providers "to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital" for AI infrastructure – with NVIDIA optionally providing "limited residual-value support" on specific projects. These are preliminary arrangements and may not become definitive. Second, a new business model with select AI cloud partners where NVIDIA commits to buy unsold capacity: six-year commitments totaling $36 billion as of 2026-07-26, shrinking as third parties use the capacity. The matching risk factor makes the obligation explicit: NVIDIA guarantees capacity purchases if unsold to third parties. This ties directly to the "circular financing" worry – NVIDIA capital funding its own demand – already a live watch item after the reported ~$250B OpenAI financing/compute deal drove a ~4% one-day drop on 2026-07-27. The 10-Q doesn't confirm or deny the OpenAI deal specifically; there's no exact revenue-at-risk figure from these guarantees, but the direction is clear: contingent obligations are stacking on top of a reported $108.5B total maximum guarantee exposure, including $105B to SB Energy Corp. for land, power and shell buildout.
3. China: the reversal (HIGH). Two risk-factor changes: PRC government restrictions prevented H20 sales despite U.S. government licenses, reversing the prior expectation of a licensed China revenue stream; and the specific China antitrust preliminary finding (alleged violation of Mellanox acquisition commitments) was replaced with generalized export-control language. Management's Q3 FY2027 guidance already excludes any Data Center compute revenue from China – guided, not reported – so the base case is consistent, but the optionality is now gone rather than deferred. New risk factors add that export rules may trigger foreign government inquiries and that overseas governments may direct customers to competitors.
Supply-side risk has eased (CoWoS capacity, HBM multi-sourcing) while demand-side concentration has risen — fewer, larger customers with growing in-house alternatives and NVDA capital entanglement.
Bottom line: The bottleneck has shifted from making chips (TSMC CoWoS, largely fixed) to selling them — watch hyperscaler capex commentary and custom-ASIC wins, not supplier capacity, as the 2026 swing factor.
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.
## Manufacturing (the single biggest chokepoint) - TSMC: ~90%+ of NVDA's advanced-node production (4nm/5nm Hopper, 4NP Blackwell). Also ~60–70% of global CoWoS advanced packaging capacity, which has been the binding constraint on AI GPU supply since 2023. - What changed: TSMC roughly doubled CoWoS capacity through 2024–2025 (~35k → ~75–80k wafers/month), easing the Blackwell ramp. NVDA has begun qualifying Samsung and Intel for packaging, but this is marginal (<10%). Taiwan geopolitical risk remains the single largest tail risk in the entire AI trade.
## Manufacturing (the single biggest chokepoint) - TSMC: ~90%+ of NVDA's advanced-node production (4nm/5nm Hopper, 4NP Blackwell). Also ~60–70% of global CoWoS advanced packaging capacity, which has been the binding constraint on AI GPU supply since 2023. - What changed: TSMC roughly doubled CoWoS capacity through 2024–2025 (~35k → ~75–80k wafers/month), easing the Blackwell ramp. NVDA has begun qualifying Samsung and Intel for packaging, but this is marginal (<10%). Taiwan geopolitical risk remains the single largest tail risk in the entire AI trade.
## Customer concentration (equally important) - Hyperscalers = ~50%+ of data center revenue. Microsoft, Meta, Amazon, Google each likely 8–13% of total revenue. - What changed: Two big shifts in 2025: 1. Custom silicon competition – Google TPU (Broadcom), Amazon Trainium are winning inference share; Meta/Google buying AMD MI300X as second source. 2. Circular financing – NVDA's ~$100B OpenAI commitment and equity stakes in CoreWeave, xAI-linked deals raise questions about demand durability and round-tripping. 3. China: export controls cut China from ~20–25% of DC revenue historically to low-single digits; the H20 ban (April 2025) forced a ~$4.5B write-down, partially reversed later under a 15% revenue-share deal with the US government – policy is now a live revenue variable.
## Customer concentration (equally important) - Hyperscalers = ~50%+ of data center revenue. Microsoft, Meta, Amazon, Google each likely 8–13% of total revenue. - What changed: Two big shifts in 2025: 1. Custom silicon competition – Google TPU (Broadcom), Amazon Trainium are winning inference share; Meta/Google buying AMD MI300X as second source. 2. Circular financing – NVDA's ~$100B OpenAI commitment and equity stakes in CoreWeave, xAI-linked deals raise questions about demand durability and round-tripping. 3. China: export controls cut China from ~20–25% of DC revenue historically to low-single digits; the H20 ban (April 2025) forced a ~$4.5B write-down, partially reversed later under a 15% revenue-share deal with the US government – policy is now a live revenue variable.
## Manufacturing (the single biggest chokepoint) - TSMC: ~90%+ of NVDA's advanced-node production (4nm/5nm Hopper, 4NP Blackwell). Also ~60–70% of global CoWoS advanced packaging capacity, which has been the binding constraint on AI GPU supply since 2023. - What changed: TSMC roughly doubled CoWoS capacity through 2024–2025 (~35k → ~75–80k wafers/month), easing the Blackwell ramp. NVDA has begun qualifying Samsung and Intel for packaging, but this is marginal (<10%). Taiwan geopolitical risk remains the single largest tail risk in the entire AI trade.
NVDA is fabless — it owns no fabs — so its supply chain is its concentration risk. The exposure sits in three layers: inputs (foundry and memory), assembly (ODMs), and demand (a handful of direct customers, mostly ODMs and hyperscalers). Here's the map, then what changed.
The FY2026 10-K names Samsung in both the foundry and memory sentences, and the live signals are adverse on memory. Samsung and SK Hynix inventories have dropped below 10 days — a critical HBM/DRAM shortage that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" [reported headline + relationship evidence; medium confidence]. This is corroborated by a second headline on 2026-09-07 warning of a "hard capacity ceiling near-term." The watch item is whether Samsung allocates to NVDA vs competitors.
That shortage connects directly to management's own numbers: the Q4 FY2027 gross margin reset to 71–72% is guided on "memory cost inflation" [earnings recap]. Gross margin printed [state]">75.0% in Q2 FY2027 (quarter ended July 26, 2026) [Q2 FY2027], so this is a real input-cost pass-through, not a modeling hypothetical. NVDA being fabless means the shock hits its COGS through supplier pricing — which is exactly what's happening.
Foundry dependence is a listed high risk factor in its own right: "Dependence on third-party foundries reduces control over product quantity, quality, yields, and delivery schedules" [risk]. No specific foundry is named in the reference beyond Samsung, so I can't quantify TSMC exposure — but single-source dependence is structural.
The ODM layer looks healthy. Hon Hai (Foxconn), the largest assembly partner, reported record August sales and +52% growth with European AI orders expanding — "strong evidence that NVIDIA's largest assembly partner is capturing massive AI server volumes; supply chain flowing" [reported headline + relationship evidence; medium confidence]. Wistron raised $1.47B via GDR to fund AI server production ramp-up, reducing supply-constraint risk [reported headline + relationship evidence; medium confidence]. Note the filing's own caveat: Hon Hai sits on BOTH sides of the chain — supplier, and a leading candidate for the anonymized large direct customers.
The Q2 FY2027 10-Q shows NVDA locked down supply at unprecedented scale:
To fund this, NVDA issued [10-Q notes]">$25.0B of senior unsecured notes in June 2026; total debt went from $8.5B to $33.4B in the quarter, against [Q2 FY2027]">$22.4B cash [Q2 FY2027]. Non-marketable securities rose to [10-Q notes]">$51.2B from [10-Q notes]">$22.3B at FY-end — more strategic equity stakes [10-Q notes]. So the balance sheet itself is becoming part of the demand-financing chain.
The listed high risk factor says it plainly: significant revenue stems from limited partners, and losing one "could adversely affect revenue" [risk]. CoreWeave's CDS — the financing canary for the neocloud layer — recovered from ~[watch]">881bps to ~[watch]">452bps; back above [watch]">600bps signals renewed stress [watch].
Adverse: memory shortage (under-10-day inventories) now feeding into a guided margin reset. Beneficial: assembly partners at record volumes, geographic diversification via Europe, and demand so strong customers prepay. Structural: NVDA has answered the shortage by more than doubling supply commitments 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 and extending $105B of credit guarantees — securing supply, but also tying its balance sheet ever tighter to its customers' buildouts.
Also disclosed, not central here: inventory provisions of $784M in Q2 (H1 $1.6B), zero accrued legal contingencies, and $99.3B remaining buyback authorization [10-Q notes].
What to watch: (1) Samsung/SK Hynix allocation announcements — does NVDA win the memory share its [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 commitments imply; (2) Q4 FY2027 gross margin landing inside the guided 71–72% — below that, the reset is deeper than memory alone; (3) CoreWeave CDS back above [watch]">600bps; (4) confirmation/denial of the ~$250B OpenAI deal structure [watch].
Bottom line: demand and assembly are fine; the real concentration risk is memory supply on the input side and customer-financing entanglement on the output side — both now visible in the filings, not just the headlines.
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 FY2026 10-K names Samsung in both the foundry and memory sentences, and the live signals are adverse on memory. Samsung and SK Hynix inventories have dropped below 10 days – a critical HBM/DRAM shortage that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" . This is corroborated by a second headline on 2026-09-07 warning of a "hard capacity ceiling near-term." The watch item is whether Samsung allocates to NVDA vs competitors.
NVDA is fabless – it owns no fabs – so its supply chain is its concentration risk. The exposure sits in three layers: inputs (foundry and memory), assembly (ODMs), and demand (a handful of direct customers, mostly ODMs and hyperscalers). Here's the map, then what changed.
Adverse: memory shortage (under-10-day inventories) now feeding into a guided margin reset. Beneficial: assembly partners at record volumes, geographic diversification via Europe, and demand so strong customers prepay. Structural: NVDA has answered the shortage by more than doubling supply commitments to $279B and extending $105B of credit guarantees – securing supply, but also tying its balance sheet ever tighter to its customers' buildouts.
- One direct customer was 16% of Q2 FY2027 revenue; for H1, three customers were 16%, 15%, and 13% – all Compute & Networking . - Five direct customers held 22%, 14%, 13%, 11%, 10% of the $63.1B receivables balance . DSO stretched from ~45 days in Q1 FY2027 to ~60 days in Q2 – slower collections as the customer base concentrates Q2 FY2027 Q1 FY2027. - 38% of Q2 revenue came from customers headquartered outside the US, up from 30% a year ago . - Customer advances (money prepaid before delivery) doubled: $15.6B in H1 FY2027 vs $7.5B a year ago – customers are paying up front to secure allocation, which is what a shortage market looks like.
- One direct customer was 16% of Q2 FY2027 revenue; for H1, three customers were 16%, 15%, and 13% – all Compute & Networking . - Five direct customers held 22%, 14%, 13%, 11%, 10% of the $63.1B receivables balance . DSO stretched from ~45 days in Q1 FY2027 to ~60 days in Q2 – slower collections as the customer base concentrates Q2 FY2027 Q1 FY2027. - 38% of Q2 revenue came from customers headquartered outside the US, up from 30% a year ago . - Customer advances (money prepaid before delivery) doubled: $15.6B in H1 FY2027 vs $7.5B a year ago – customers are paying up front to secure allocation, which is what a shortage market looks like.
The FY2026 10-K names Samsung in both the foundry and memory sentences, and the live signals are adverse on memory. Samsung and SK Hynix inventories have dropped below 10 days – a critical HBM/DRAM shortage that "threatens NVIDIA's ability to complete GPU modules; pricing pressure likely" . This is corroborated by a second headline on 2026-09-07 warning of a "hard capacity ceiling near-term." The watch item is whether Samsung allocates to NVDA vs competitors.
- Supply and capacity commitments jumped from $119.0B (as of 2026-04-26) to $279.0B as of July 26, 2026 – more than doubled in one quarter, "to secure the necessary supply and critical components needed to meet demand for the next several years" . That's nearly 3x FY2026's reported revenue of $215.9B. - Total future purchase and other commitments: $366.0B, with $120.0B due in the rest of FY2027 and $100.0B next year . - On top: $36.0B in AI cloud agreement commitments, $29.0B cloud service agreements, $25.0B data center leases not yet commenced, and $25.0B equity investment commitments . - Guarantees: in August 2026 NVDA entered guarantees capped at $105.0B with affiliates of SB Energy Corp. "to provide credit support on a land, power, and shell buildout" – NVDA is now credit-supporting its customers' data centers. Total guarantee maximum exposure $108.5B . This is the "circular financing" concern the openaicircularfinancing watch item tracks: NVDA capital funding its own demand .
The ODM layer looks healthy. Hon Hai (Foxconn), the largest assembly partner, reported record August sales and +52% growth with European AI orders expanding – "strong evidence that NVIDIA's largest assembly partner is capturing massive AI server volumes; supply chain flowing" . Wistron raised $1.47B via GDR to fund AI server production ramp-up, reducing supply-constraint risk . Note the filing's own caveat: Hon Hai sits on BOTH sides of the chain – supplier, and a leading candidate for the anonymized large direct customers.
Caveat up front: my data runs to early 2025, so I can speak to the pattern from the most recent 10-Q/earnings-call pairs I have (through FY25) — verify against the filing you're actually holding. Here's what consistently shows up in the 10-Q that management does not emphasize on the call, using the recent quarters as the template:
1. Customer concentration and receivables risk
2. Exploding purchase commitments and inventory
3. Export control exposure quantified
4. Related-party and circular-flow details
Which ones change how you read the guidance?
Bottom line: Read guidance off the 10-Q's purchase-commitment and customer-concentration tables, not the call's demand narrative — the filing shows NVDA is paying for its own runway with a very narrow set of payers. Confirm against the current filing, as my figures reflect data through early 2025.
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 10-Q filed 2026-08-26 quietly disclosed three things management did not put on the call that same day: a set of MOUs to mobilize $500 billion of third-party AI-infrastructure capital, a new business model under which NVIDIA commits to buy unsold capacity from AI cloud partners ([Q3 FY2026]">$36 billion outstanding), and — just after quarter-end — $105 billion of guarantees to SB Energy Corp. Those three, more than anything on the call, change how the [guidance]">$108 billion Q3 FY2027 revenue guidance should be read.
1. The $500B financing-platform MOUs. In August 2026 NVIDIA signed memorandums with large capital providers to "mobilize more than $500 billion of third-party capital over time," with NVIDIA optionally providing "limited residual-value support" on some projects [SEC filings]. The call's tone was demand-conviction; the filing reveals demand increasingly needs financing architecture NVIDIA helps arrange. The filing itself cautions these "may not lead to definitive agreements."
2. The AI cloud capacity backstop — a new business model. NVIDIA introduced commitments, "typically six years in duration," to purchase capacity from select AI cloud partners if third-party customers don't use it — [Q3 FY2026]">$36 billion as of July 26, 2026, shrinking as capacity gets consumed [SEC filings]. This is a contingent demand guarantee: if AI-startup/model-builder demand disappoints, NVIDIA becomes the buyer of last resort. A companion risk factor — guarantees of AI cloud capacity purchases if unsold — is entirely new [SEC filings].
3. The SB Energy guarantees — $105B, signed after quarter-end. Guarantees capped at $105 billion to credit-support a "land, power, and shell buildout" with SB Energy affiliates, entered August 2026 — plus [10-Q notes]">$3.5B maximum exposure under earlier AI-cloud land/power/shell guarantees [10-Q notes]. Total maximum guarantee exposure: [10-Q notes]">$108.5 billion [10-Q notes]. Nothing this size was flagged on the call.
4. Supply commitments more than doubled — $119B to $279B in one quarter [10-Q notes]. Management framed this as securing supply "to meet demand," and it matches the Rubin ramp. But it is also [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 of largely non-cancellable purchase obligation against a guided ~[guidance]">$108B quarter — and the filing pairs it with new language that customers may postpone purchases for lack of data-center infrastructure or capital [SEC filings].
5. China disclosures shifted. The prior quarter's specific antitrust preliminary finding (alleged Mellanox-commitment violations) was replaced with generalized export-control language, and a new risk factor states PRC restrictions prevented H20 sales despite U.S. licenses — NVIDIA could not sell inventory it had built [SEC filings]. Also new: export rules may trigger foreign-government inquiries, and foreign governments may steer customers to competitors [SEC filings].
6. A standalone indebtedness risk factor — new, covering debt service and covenant restrictions, after NVIDIA issued [Q2 FY2027]">$25 billion of senior unsecured notes in June 2026 (reported), lifting total debt to [Q3 FY2026]">$33.4 billion from [Q1 FY2026]">$8.5 billion one quarter earlier [10-Q notes].
Management guided Q3 FY2027 (quarter ending October 2026) revenue of [guidance]">$108.0B ±[guidance]">2% and GAAP/non-GAAP gross margin of [guidance]">74.0% ±50bps — guided figures, explicitly excluding any Data Center compute revenue from China [guidance].
Also disclosed, not material to this question: the investment portfolio shifted to $103.2B total with [10-Q notes]">$36.9B under lock-up and $12.5B of H1 unrealized equity gains; no legal loss accrual ([10-Q notes]">$0); $99.3B of buyback authorization remaining [10-Q notes].
Bottom line: The call sold certainty; the 10-Q shows NVIDIA is increasingly financing the certainty — [Q3 FY2026]">$36B of capacity backstops, $105B of energy guarantees, $500B of MOU capital, 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 bets. The [guidance]">$108B guide is achievable, but read it as demand NVIDIA is underwriting, not merely observing.
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.
4. Supply commitments more than doubled – $119B to $279B in one quarter . Management framed this as securing supply "to meet demand," and it matches the Rubin ramp. But it is also $279 billion of largely non-cancellable purchase obligation against a guided ~$108B quarter – and the filing pairs it with new language that customers may postpone purchases for lack of data-center infrastructure or capital .
The 10-Q filed 2026-08-26 quietly disclosed three things management did not put on the call that same day: a set of MOUs to mobilize $500 billion of third-party AI-infrastructure capital, a new business model under which NVIDIA commits to buy unsold capacity from AI cloud partners ($36 billion outstanding), and – just after quarter-end – $105 billion of guarantees to SB Energy Corp. Those three, more than anything on the call, change how the $108 billion Q3 FY2027 revenue guidance should be read.
1. The $500B financing-platform MOUs. In August 2026 NVIDIA signed memorandums with large capital providers to "mobilize more than $500 billion of third-party capital over time," with NVIDIA optionally providing "limited residual-value support" on some projects . The call's tone was demand-conviction; the filing reveals demand increasingly needs financing architecture NVIDIA helps arrange. The filing itself cautions these "may not lead to definitive agreements."
- The China exclusion is not just conservatism – it's forced. The filing's H20 disclosure shows the PRC blocked licensed sales. Read the "conservative assumptions" framing as a hard zero, not optionality likely to appear. - The revenue guide leans on financed demand. Between the $36B capacity backstop, the $500B MOUs, and customer advances inside deferred revenue ($15.6B of $17.7B in H1 additions, reported) , a growing share of demand is capital-structured. If credit conditions for AI clouds tighten, the guide's ±2% band understates the tail. - Margin guide has its own footnote. Q2 reported gross margin was 75% Q2 FY2027, but the filing shows $784 million of inventory provisions in Q2 ($1.6B for H1) , and with supply commitments at $279B, any demand wobble turns into write-downs against that 74% guide. Management has already guided Q4 FY2027 gross margin down to 71–72% – treat the 74% Q3 guide as a peak, not a plateau. - Concentration cuts both ways. One direct customer was 16% of Q2 revenue; three were 16%, 15%, 13% of H1 (reported) . Guidance risk is a handful of purchase decisions, plus five customers holding 22%, 14%, 13%, 11%, 10% of receivables – against DSO that stretched to ~60 days from ~45 in Q1 Q2 FY2027.
5. China disclosures shifted. The prior quarter's specific antitrust preliminary finding (alleged Mellanox-commitment violations) was replaced with generalized export-control language, and a new risk factor states PRC restrictions prevented H20 sales despite U.S. licenses – NVIDIA could not sell inventory it had built . Also new: export rules may trigger foreign-government inquiries, and foreign governments may steer customers to competitors .
6. A standalone indebtedness risk factor – new, covering debt service and covenant restrictions, after NVIDIA issued $25 billion of senior unsecured notes in June 2026 (reported), lifting total debt to $33.4 billion from $8.5 billion one quarter earlier .
The 10-Q filed 2026-08-26 quietly disclosed three things management did not put on the call that same day: a set of MOUs to mobilize $500 billion of third-party AI-infrastructure capital, a new business model under which NVIDIA commits to buy unsold capacity from AI cloud partners ($36 billion outstanding), and – just after quarter-end – $105 billion of guarantees to SB Energy Corp. Those three, more than anything on the call, change how the $108 billion Q3 FY2027 revenue guidance should be read.
- The China exclusion is not just conservatism – it's forced. The filing's H20 disclosure shows the PRC blocked licensed sales. Read the "conservative assumptions" framing as a hard zero, not optionality likely to appear. - The revenue guide leans on financed demand. Between the $36B capacity backstop, the $500B MOUs, and customer advances inside deferred revenue ($15.6B of $17.7B in H1 additions, reported) , a growing share of demand is capital-structured. If credit conditions for AI clouds tighten, the guide's ±2% band understates the tail. - Margin guide has its own footnote. Q2 reported gross margin was 75% Q2 FY2027, but the filing shows $784 million of inventory provisions in Q2 ($1.6B for H1) , and with supply commitments at $279B, any demand wobble turns into write-downs against that 74% guide. Management has already guided Q4 FY2027 gross margin down to 71–72% – treat the 74% Q3 guide as a peak, not a plateau. - Concentration cuts both ways. One direct customer was 16% of Q2 revenue; three were 16%, 15%, 13% of H1 (reported) . Guidance risk is a handful of purchase decisions, plus five customers holding 22%, 14%, 13%, 11%, 10% of receivables – against DSO that stretched to ~60 days from ~45 in Q1 Q2 FY2027.
Also disclosed, not material to this question: the investment portfolio shifted to $103.2B total with $36.9B under lock-up and $12.5B of H1 unrealized equity gains; no legal loss accrual ($0); $99.3B of buyback authorization remaining .
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 | Finbly research |
|---|---|---|
| Checklist matchescontextual text matches, with the supporting passage retained; not factual accuracy | 25% | 90% |
| Numbers in saved evidencescalar occurrences, not verification of the associated claims | 41 | 291 |
| Claim accuracynot assessed by the numeric occurrence checker | Not assessed | Not assessed |
| Numbers without a scalar matchno scalar match in the saved packet, or an ambiguous range; not an error count | 93 | 40 |
| Research reference tagscategory labels in the answer; repeats counted | 0 | 154 |
| Input tokens per questionwhat the model read | 180 | 9.5k |
| Cost per questionmodel list prices; a dash means no published rate | $0.0097 | $0.043 |
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