A First Full-Year Guide at Roughly 70% Growth, Paid For With 350 Basis Points of Gross Margin and a $27 Billion Cash-Flow Swing

Results vs. Consensus

Q2 FY2027 Scorecard

Year-Over-Year Comparisons

Quarter-Over-Quarter Comparisons

Quality of Beat

Revenue. Revenue more than doubled year-over-year and growth accelerated for a fourth consecutive quarter, which at a $96.2B quarterly base is the more remarkable of the two facts. The composition is healthier than the headline: ACIE grew 25% sequentially against Hyperscale's 13%, sovereign revenue grew 35% sequentially and more than tripled year-over-year, and the neocloud partner base is expected to exit the year at 8 gigawatts of installed capacity against roughly 3 gigawatts at the end of 2025.

Margins. We have tracked the mid-seventies framework since our initiation, when management downgraded its language from a delivered result to an effort and we called it the most important unpriced risk. It then held for three consecutive quarters and we recorded that we had been wrong on direction. This quarter the risk arrived. The framework is gone, replaced by a specific path with a named trough, and the cause is the one originally suspected. We take no satisfaction in the call: the timing was wrong by a year, which in practice means it was wrong.

EPS and cash. The income statement and the cash flow statement tell different stories this quarter and both are true. Operating margin expanded to 66.5% while free cash flow conversion fell to 22.2%, because the difference between them is working capital: inventory up 22% sequentially ahead of the Vera Rubin launch, and receivables at $63.1B on payment terms extended to large customers. Neither is a loss. Both are cash that has left the business ahead of revenue that has not yet arrived.

Segment Performance

Revenue by Market Platform

Hyperscale: $48.7B, 54.7% of Data Center

Hyperscale more than doubled year-over-year and grew 13% sequentially on sustained Blackwell strength. Management made a point it has been building toward for three quarters, that the customers deploying its compute are themselves performing better: hyperscale customers delivered accelerating revenue growth and expanding margins in the quarter. The forward capital expenditure framing is now very large indeed, with cloud industry backlog above $2 trillion and top-five hyperscaler capital expenditure expected to reach nearly $800B in 2026 and $1.3 trillion in 2027.

The single largest commercial announcement was an AWS expansion: an additional 2 million GPUs deploying from this quarter through Q2 fiscal 2029, alongside Vera CPUs both integrated with Rubin and standalone, with AWS also serving NVIDIA's open model family and adopting the full physical AI stack for its warehouse robot fleet.

Assessment: Hyperscale growing 13% sequentially is the slower half by design, and management guided it to reaccelerate in Q4 and into fiscal 2028 as Vera Rubin supply grows. The customer-profitability point is the one that matters for the durability question. Compute that expands the buyer's own margins is compute that gets bought again.

ACIE: $40.3B, Growing 25% Sequentially and 138% Year-Over-Year

ACIE grew 138% year-over-year on neocloud capacity additions serving enterprises, AI startups and sovereigns, plus hyperscalers purchasing supplementary capacity through AI clouds. Sovereign revenue, routed primarily through regional neoclouds, grew 35% sequentially and more than tripled year-over-year. The named neocloud roster now spans Armenia, Africa, Taiwan, India, Australia and Malaysia alongside the established operators.

Management's structural argument is that a country or region will allocate land and power to a regional cloud partner in ways it would not to a foreign hyperscaler, and that NVIDIA's neutrality as a non-operator of clouds is what makes it the natural partner.

Assessment: The growth is real and the reported share is not what it appeared last quarter. The recast moved roughly $5.18B from ACIE to Hyperscale, which means the crossover we flagged as imminent has been pushed out by a reclassification rather than by performance. That is a fair accounting change and it is also a reminder that segment-level conclusions built on one quarter of a new framework are fragile. The one that survives is the growth differential: 25% against 13%, in the same direction for a second consecutive quarter.

Edge Computing: $7.2B, With Memory Prices Visible in the Consumer Line

Edge Computing grew 27% year-over-year and 13% sequentially on strong Blackwell workstation sales, partially offset by slower consumer personal computer sales tempered by elevated memory and system prices. This is the second consecutive quarter with the same explanation.

Assessment: The consumer softness is a useful independent read on the memory situation. The same input-cost pressure that is about to compress Data Center gross margin is already visible as a demand effect at the edge, which suggests the pricing environment management described is broad rather than specific to high-bandwidth memory procurement.

Networking, Now Inside Data Center

Networking had another record quarter with revenue growing 18% sequentially, and Spectrum-X Ethernet grew 2.6 times year-over-year. Management repeated the claim that NVIDIA is becoming the largest and fastest-growing networking company in the world. Under the new framework there is no standalone networking figure, so the growth rate is all investors receive.

Assessment: A 2.6x year-over-year Ethernet growth rate is a deceleration from the 199% total networking growth disclosed one quarter ago on the old basis, though the two are not directly comparable. This is the first quarter with no standalone networking figure at all, the prior quarter having disclosed it one final time under the legacy sub-markets, and the visibility loss is now material. We continue to ask for its reinstatement.

Key Topics & Management Commentary

Overall Management Tone: Deliberately transparent to an extent that is new, and the transparency ran in both directions. Management volunteered a first-ever full-year guide, pre-empted the gross margin question in prepared remarks rather than waiting for it, and named the circular-financing critique itself before any analyst raised it. That is a management team that has decided its disclosure posture is a competitive variable. The tone was least convincing on customer payback economics, where the supporting evidence offered for a central claim was second-hand.

1. The First Full-Year Guide in Company History

NVIDIA has never guided a fiscal year in advance. It did so here, and framed the number as a supply ceiling rather than a demand forecast.

"We expect to grow revenue by approximately 70% in fiscal 2028. This is a supply-constrained outlook."
— Colette Kress, Executive Vice President and Chief Financial Officer

Management said customer forecasts point to growth roughly doubling, and that supply permits approximately 70%. Asked what gives the confidence to guide a year out after never having done so, the CEO's answer was about supply-chain visibility earned by working further upstream and downstream than a chip company normally would, and about wanting the whole ecosystem to plan against one set of numbers.

"It is the case that we've never forecasted or never guided to a year in advance. And even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The Street was modelling 44%. A 26-point gap on a company this size is the largest single revision we have seen in this coverage period, and it is backed by $279B of supply commitments rather than by a slide. The framing matters as much as the number: a guide described as supply-constrained can be missed only through execution failure, not through demand disappointment, which is a fundamentally different risk to underwrite.

2. The Margin Framework Is Reset, and Disclosed Unprompted

The mid-seventies framework, carried since our initiation, was abandoned in prepared remarks. Management named the cause, gave a specific path with a trough, and explained why it was raising the subject itself.

"We expect margins to bottom in Q4 in the 71% to 72% range before settling at 72% to 73% in fiscal year '28, as executed price increases take effect in Q1. We want to be direct about this rather than let it linger as an open question."
— Colette Kress, Executive Vice President and Chief Financial Officer

The stated driver is memory, described as extreme pricing conditions whose magnitude has exceeded prior expectations and is heading higher into next year. Management offered a framing that is more than a cost complaint: memory scarcity is itself a symptom of the AI build-out driving NVIDIA's own growth, which distinguishes it from a component whose price rises with no offsetting benefit.

Assessment: The reset is roughly 350 basis points from the 75.0% just delivered to the 71.5% Q4 trough, which on the guided Q4 revenue base is a large absolute number. Two things soften it. The price increases described as executed rather than planned mean the fiscal 2028 recovery to 72% to 73% has a mechanism rather than a hope. And a company that pre-empts its own worst disclosure has earned some credit on the next one. Neither changes the arithmetic: this is the risk we identified at initiation, arriving a year later than we expected.

3. Cash Conversion Broke, and the Reasons Are Legible

Free cash flow fell to $21,341M from $48,554M, a $27,213M sequential decline, taking conversion to 22.2% of revenue from 59.5%. Operating cash flow fell 52% to $24.1B on higher working capital adjustments and cash taxes, the latter of which management had pre-announced last quarter.

The receivables movement is the item that deserves attention. Days sales outstanding rose to 60 days from 45, which management attributed to extended payment terms on large multi-quarter agreements with certain investment-grade customers. Last quarter management guided that days sales outstanding would return to the mid-fifties from 45. It went to 60 instead.

Assessment: Three things are happening at once and only one of them is concerning. Cash taxes were flagged in advance and are timing. Inventory building 22% ahead of a Vera Rubin launch is correct behaviour. Receivables extending to 60 days on granted payment terms is the one that changes the character of the revenue: NVIDIA is now financing part of its own sales cycle, and it missed its own forecast on the metric by five days in the wrong direction. That is the number to watch next quarter.

4. NVIDIA Names the Circular-Financing Critique Itself

The most striking passage of the call was management raising the bear case against itself before any analyst did.

"We recognize the scale of this support, and we know some will call this circular financing. We see it differently."
— Colette Kress, Executive Vice President and Chief Financial Officer

The scale disclosed is substantial. NVIDIA has invested nearly $50B in frontier AI labs. It has partnered with six infrastructure capital providers to establish financing platforms intended to raise over $500B of third-party capital. It has secured land, power and shell capacity for a 4.25 gigawatt campus that will exclusively host NVIDIA compute under twenty-year leases to OpenAI, with guarantee obligations capped at $105B, and holds an option to support roughly 3.8 additional gigawatts at the same site. OpenAI's existing and planned commitments represent approximately 12 gigawatts. For a second frontier lab, NVIDIA will provide selective credit enhancement for nearly 2 gigawatts.

The defence rests on three claims: the labs are constrained by compute rather than by technology or customer demand, the NVIDIA platform is fungible and durable so it can be redeployed if a customer fails, and the compute shipped will be consumed by investment-grade customers or by parties backed by one. Management also quantified the exposure.

"we expect demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year."
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: This is the disclosure we have asked for across three consecutive quarters, and it is larger than we expected. Roughly a quarter of fiscal 2028 revenue depending on customers whose purchases lean on NVIDIA's balance sheet is a material concentration of a new kind. The fungibility argument is genuinely strong and is the reason this is not a subprime analogy: if a lab fails, the racks are redeployable to a market that is short of them. But the honest characterisation is that NVIDIA has moved from selling compute to underwriting the demand for compute, and those carry different multiples.

5. Revenue Per Gigawatt Rises to $40 Billion

The clearest quantification of the platform strategy is the revenue opportunity per gigawatt of data center capacity, which management laid out generationally.

"In the case of Hopper, we were at about $18 billion per gigawatt. For Grace Blackwell, we're about $25 billion per gigawatt. And for Vera Rubin, it's about $40 billion per gigawatt."
— Jensen Huang, Founder and Chief Executive Officer

The Vera Rubin figure now spans Vera CPU, Rubin GPU, NVLink, InfiniBand or Ethernet, and the newly productized decode accelerator. Asked whether the progression continues beyond Vera Rubin, management said it would go higher and framed the objective plainly: put as much compute as possible into each gigawatt of land and power.

Assessment: This is the single most useful modelling metric the company discloses, because it converts the gigawatt figures that dominate AI infrastructure discussion into NVIDIA revenue. It also reframes the ASIC threat: a competitor winning the accelerator socket contests perhaps half of a $40B-per-gigawatt opportunity, not all of it. The content expansion from $18B to $40B across two generations is the platform strategy working.

6. Vera Rubin in Full Production, at 20% of Q3 Data Center Revenue

Production shipments of Vera Rubin commenced in the month of the call, with purchase orders already received from every major hyperscaler, AI cloud and system original equipment manufacturer. Management expects it to be the fastest product ramp in company history and, for the first time, sized the near-term contribution: approximately 20% of Data Center revenue in Q3. The platform delivers 30 times higher throughput per megawatt and 35 times lower token costs relative to Grace Blackwell Ultra.

Assessment: After three quarters of declining to size the ramp, management has given a number, and 20% of Q3 Data Center revenue in the first full quarter is a fast start. This retires the largest modelling uncertainty we have carried. The residual question shifts to whether the fastest-ramp claim survives the margin trough, since Q4 is both the peak Rubin ramp quarter and the guided gross margin low.

7. The Decode Accelerator Ships, and the Story Changed

One quarter after management characterized the acquired low-latency decode technology as a niche product with an insignificant customer count, the resulting rack-scale system is in full production, is claimed by management to deliver nearly four times the tokens per second of the next best alternative on the benchmark it cited, is included in the $40B-per-gigawatt content figure, and ships in volume later this quarter with a named lead customer.

Management's positioning remains that the vast majority of data centers will run Vera Rubin and NVLink 72, with the decode accelerator bolted on for services requiring very high interactivity at higher cost per token and higher average selling prices.

Assessment: The characterization has moved a long way in ninety days, from niche with few customers to full production, record benchmarks and inclusion in the headline content metric. Both statements can be true if the addressable slice is small but the attach economics are good. We note the shift rather than resolve it, and we continue not to model a separate revenue line for it. The $13.0B paid two quarters ago looks better today than it did last quarter.

8. Commitments Reach $366 Billion and Guarantees Reach $108.5 Billion

Supply and capacity commitments more than doubled to $279B from $119B, which management attributed primarily to memory procurement. Adding cloud service agreements of $29B, data center leases not yet commenced of $25B, equity investments of $25B and capital expenditures of $8B brings that table's total to $366B, of which $120B falls in the remainder of fiscal 2027 and $100B in fiscal 2028. A second table, headed Additional Commitments, carries a further $56B: $36B of AI cloud agreements and $20B of third-party data center leases not yet commenced.

Separately, maximum gross guarantee exposure is $108.5B: $3.5B of land, power and shell guarantees for AI cloud partners, plus $105.0B for the OpenAI-linked campus, phased in as data centers become ready for service with the first expected in fiscal 2029 and declining as lease payments are made.

NVIDIA also issued $25.0B of senior unsecured notes in the quarter for general corporate purposes.

Assessment: The commitment book now exceeds the $303B of revenue NVIDIA has generated over the trailing twelve months. The memory attribution ties it directly to the margin reset, which makes both disclosures more credible for being consistent. The guarantee structure is the genuinely new instrument: it is contingent, phased, capped and declining, which is materially better than an unfunded obligation, and it is still $105B of exposure to a single counterparty's lease performance. A company issuing $25B of debt while generating $21B of quarterly free cash flow is making a deliberate choice to fund commitments with the balance sheet rather than from operations.

9. Capital Return Runs Ahead of the Commitment

NVIDIA returned a record $25.8B in the quarter, comprising $19.7B of repurchases and $6.0B of dividends at the raised $0.25 rate, and ended with approximately $99.0B remaining under authorization. Management reported tracking ahead of the payout framework it introduced last quarter, at 60% of free cash flow year-to-date against a plan of 50% or more, and said it intends to return excess free cash flow net of strategic uses going forward.

Assessment: The commitment we said was missing two quarters ago is not only in place but being exceeded and reported against. That is the correct behaviour and it deserves to be said plainly. The qualifier "net of strategic uses" is doing real work in that sentence, and given the scale of the strategic uses now disclosed, it is the clause that determines what the framework actually means.

10. Demand Evidence: Strong on Aggregates, Thin on Payback

Management supplied unusually specific demand aggregates. Global venture funding in AI, roughly 70% of which is spent on compute, exceeded $400B in the first half of 2026 against $265B for all of 2025. Nearly 20 companies now exceed $1B in annualized run-rate revenue, up from 13 at the end of last year. On-premises automotive vertical revenue reached $8B on a trailing twelve-month basis, with financial services, manufacturing and healthcare combining for $7B.

On the question that actually underpins the cycle, whether customers earn a return on the compute, the evidence was weaker.

"I heard the other day, that return on invested capital is now less than a year."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The aggregates are excellent and largely externally verifiable. The payback claim is not: it was prefaced as something heard rather than measured, and it is the load-bearing assumption for the entire build-out. This is the clause that would break first if the cycle turns, and it is the one supported by the least evidence on this call. The Meta conversion datapoints cited in prior quarters were better.

11. China Reopens Slightly, and It Is Margin-Dilutive

Hopper 200 shipments to China-based customers resumed under United States licenses at less than 1% of Data Center revenue. Management disclosed something new and analytically useful: those current Hopper shipments are dilutive to corporate gross margins. Given ongoing geopolitical uncertainty there is still no China Data Center compute revenue in the forward outlook.

Assessment: We have modelled China at zero for four quarters and continue to. The new information changes the shape of the optionality rather than its size. A reopening at Hopper-generation margins would add revenue and subtract from the corporate gross margin at exactly the moment margin is already compressing, which makes it a mixed event rather than the pure upside the market has treated it as.

Guidance & Outlook

Two guides were issued and they point in opposite directions. The Q3 revenue guide of $108.0B implies 12.2% sequential growth and roughly 89% growth against the $57,006M printed a year ago, with sequential growth driven primarily by ACIE and hyperscale expected to reaccelerate in Q4 as Vera Rubin supply grows. The fiscal 2028 guide of approximately 70% growth is the first full-year outlook in company history and lands roughly 26 points above where the Street was modelling. Against both, gross margin steps down for three consecutive quarters before recovering.

Implied Q-over-Q ramp: Applying the Q3 guide and a fourth quarter consistent with the reaccelerating hyperscale commentary puts fiscal 2027 revenue near $405B. Growing that approximately 70% implies fiscal 2028 revenue near $690B, which is the number the ~70% guide actually commits to.

Street at: Consensus had fiscal 2028 growth at 44%. The gap is not a rounding difference in a model; it is a different view of what the company can supply.

Guidance style: The pattern has now changed twice in three quarters. Management guided above the Street, then below it, and has now issued a full-year number nearly double the Street's. The unifying explanation is that management has concluded its own supply visibility is better information than the market's demand modelling, and has decided to publish it so that customers, suppliers and shareholders plan against the same figure. That is a defensible reason to break a decade of practice, and it also means the guide is now the thesis: a miss on the ~70% would be a far more serious event than any single quarterly miss in this coverage period.

Analyst Q&A Highlights

What Gives the Confidence to Guide a Full Year, and What Separates 70% From 100%

The opening question went directly to the two things the guide raised: why guide a year out after never doing so, and what constraint separates the guided 70% from the roughly 100% demand growth described.

Q: "I wonder if you could give us color on the 70%, and what gives you the confidence to guide a full year out if you haven't been doing that? And then what's the gap between that amount of growth in the 100% demand growth."
— Joseph Moore, Morgan Stanley

A: "It is the case that we've never forecasted or never guided to a year in advance. And even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Answered fully, including the part about motive. Management said it wanted customers, shareholders and the supply chain working from one set of numbers because everyone is committing resources against it. That is a coordination argument rather than a promotional one, and it is consistent with the $279B of supply commitments disclosed the same day. The answer also quietly reframed the risk: the constraint is procurement and construction lead time, including land, power and shell that is often two to three years out.

What the Growth Would Be Unconstrained

A question worked backward from the prior trillion-dollar framework to estimate the size of the upgrade, asked what drives it across the product lines, and pressed on what the number would be without the supply ceiling.

Q: "I'm also curious just it's a constrained number. What would it be if it wasn't constrained?"
— Stacy Rasgon, Bernstein Research

A: "The unconstrained would be a lot higher. It's -- we grew 100% year-over-year this year. The unconstrained is significant."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Management would not put a number on unconstrained demand, which is the right restraint. The substantive part of the answer was the revenue-per-gigawatt progression, which explains how revenue can grow faster than deployed gigawatts. The question also embedded a fair point that went unaddressed: some of the fiscal 2028 growth is price, since management confirmed executed price increases take effect in Q1. Management did not separate volume from price in the ~70%, and it should.

Balancing Ecosystem Investment Against Customers Building Rival Silicon

The sharpest question of the call totalled the disclosed commitments, asked what cash portion falls in fiscal 2028, and raised the tension in investing heavily in frontier labs that are simultaneously designing their own accelerators.

Q: "how are you balancing this dynamic where you want to invest a lot in the ecosystem, but part of that ecosystem wants to develop competitive solutions?"
— Vivek Arya, Bank of America Securities

A: "Whereas many of these XPUs are inference specific chips for one cloud or one service. NVIDIA is a platform, an entire AI factory platform that spans the entire AI life cycle that you can use in any cloud, it's in every cloud."
— Jensen Huang, Founder and Chief Executive Officer

On the investments themselves, the CEO was unusually personal, saying his only regret is not having invested more and sooner, and noting that two of the companies will likely go public soon.

Assessment: The platform-versus-chip argument is the correct rebuttal and it is not a complete one. A lab can run its own inference silicon for its own service while buying NVIDIA for everything else, which erodes the highest-volume workload without touching the platform claim. The cash-timing half of the question was answered by the CFO in terms of supply commitments being concentrated in the first three years, but the specific fiscal 2028 cash outlay for ecosystem investment was not given. That remains the one number the company still will not provide.

Ranking the Supply Constraints

A question asked management to rank the most acute constraints among data center land, power and shell availability, memory, and foundry capacity, which is the natural follow-up to a guide defined by supply.

Q: "can you maybe talk about some of the -- or rank order some of the most acute constraints, whether that be things like data center, power and shell availability, DRAM, wafer foundry availability, et cetera?"
— James Schneider, Goldman Sachs

A: "I think the answer is our entire supply chain is challenged. And it's everybody is really running flat out."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: A refusal to rank, for the second time in four quarters. It is a defensible answer and it leaves investors without the single variable to monitor. The useful detail management did give is that capacity arrives continuously rather than in steps, and that yield improvement is an ongoing lever, which means the ~70% has upside that accrues gradually rather than through a single unlock.

Whether Open Models Help or Hurt

A question probed whether the rise of open-weight models, which could take workload share from the frontier labs driving much of NVIDIA's demand, is net positive or negative for the company.

Q: "Do you see the rise of open models as being good for NVIDIA or ultimately negative?"
— Timothy Arcuri, UBS

A: "The world will need both closed models and open models."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The answer is architecturally sound and commercially self-serving in equal measure, which does not make it wrong. The specific claim worth holding is that nearly all open models run on NVIDIA because its footprint spans personal computers, edge devices, workstations, on-premises data centers and robots, and because most open models were built on NVIDIA to begin with. Open-model proliferation is a genuine hedge against frontier-lab concentration, and given that roughly a quarter of next year's business is expected to lean on lab balance sheets, that hedge is worth more than it was a year ago.

Whether Recursive Self-Improvement Inflects Demand Further

A question asked what happens to industry demand if recursive self-improvement and artificial general intelligence milestones are reached, and how management treats those as demand catalysts.

Q: "with the developments in RSI as well as AGI, what happens to industry demand? Does it inflect further?"
— Benjamin Reitzes, Melius Research

A: "We have 40,000 employees roughly. In the future, we'll have 400,000 agents, 4 million agents. And those agents are running continuously."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Management deflected the milestone framing, saying such thresholds are largely meaningless and that for many tasks they have effectively been met, and redirected to three operational claims: AI is doing useful work, it generates profitable tokens, and more compute would generate more. That is the right reframing and it returns the argument to the same load-bearing assumption as everything else on this call, which is token profitability. The agent-count framing is illustrative rather than evidentiary.

Whether Revenue Per Gigawatt Keeps Climbing Beyond Vera Rubin

The closing question asked whether the $25B to $40B per gigawatt progression continues into later generations, and what would unlock the ability to supply more.

Q: "Should we really conceptualize like 40 billion goes to 60 billion, 80 billion."
— Aaron Rakers, Wells Fargo

A: "And then eventually, with Hopper, it was 18. Now Grace Blackwell is 25. Next, Vera Rubin is 40. And after that, it's going to be higher."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Confirmed, with the important condition attached: higher content per gigawatt is only good for the customer if productivity, durability and fungibility rise alongside it, which is the discipline that keeps content expansion from becoming price gouging. The answer also contained the call's weakest evidence, the second-hand payback claim, deployed precisely where the strongest evidence was needed.

What They're NOT Saying

1. How much of the ~70% fiscal 2028 growth is price versus volume: Management confirmed executed price increases take effect in Q1 fiscal 2028 and is guiding both a revenue acceleration and a margin recovery off those increases. It did not separate the two, which makes the growth guide harder to stress-test than it should be.
2. The fiscal 2028 cash outlay for ecosystem investment: Asked directly. The answer addressed supply commitments instead. Equity investment commitments of $25B appear in the table with $18B in the remainder of fiscal 2027, but the forward cash profile of the broader programme is still not framed.
3. Which supply constraint binds: Declined to rank, for the second time in four quarters, on a guide whose entire premise is that supply is the binding constraint.
4. Any ceiling on guarantee exposure: $108.5B of maximum gross exposure is disclosed, and management holds an option to support roughly 3.8 additional gigawatts at the same campus without saying what that would add. There is no stated cap on total guarantee exposure.
5. What happens to the guarantees if a counterparty defaults: The fungibility argument covers the hardware. It does not address what NVIDIA owes on a twenty-year lease obligation for a campus whose tenant has stopped paying, which is the actual exposure.
6. Any measured evidence for customer payback: The central claim supporting the entire build-out was sourced as something management heard rather than measured. Prior quarters offered specific, named customer results. This one did not.
7. What the $2,944M "Groq, Inc." line is: A separately captioned line under that name appears in financing activities this quarter, having appeared in investing activities at $13,000M two quarters ago. Neither the call nor the CFO commentary mentions it. A transaction material enough to caption twice, in two different sections of the cash flow statement, has never been explained.
8. Standalone networking revenue, for a second quarter: A growth rate and a peer comparison were offered in place of a number.
9. Why days sales outstanding went to 60 rather than the mid-fifties management guided: The cause was described. The miss against management's own forecast was not acknowledged, and no updated forecast was given.

Market Reaction

• Pre-print setup: NVDA closed at $209.66 on August 26, up 12.4% year-to-date, up 6.7% over the trailing thirty days, but up only 15.3% over the trailing twelve months against quarterly revenue that roughly doubled in the same period. The 52-week closing range was $165.17 to $235.74.
• After-hours move: Shares rose roughly 4% on the release itself and extended the gain during the call as the fiscal 2028 guide was delivered.
• Next-day session: The stock opened at $222.86, up 6.3%, traded between $220.90 and $230.47, and closed at $227.98, up 8.7% or $18.32 from the pre-print close. The close was above the open, which had not happened in any of the three prior prints.
• Volume: 298.9 million shares against a 30-day average of 118.1 million, a 2.5x multiple, the heaviest relative to its own 30-day average of the four sessions we have covered.
• Market and peer context: The S&P 500 rose 0.7% and the Nasdaq Composite rose 1.6%. Broadcom rose 4.5% while AMD fell 0.9%.

This is the first positive reaction in our coverage of this name, and it ends a three-quarter sequence in which NVDA beat and the stock fell. The cause is not ambiguous: a first-ever full-year guide at roughly 70% against a Street at 44% is the single largest new piece of information the company has released in this period, and it arrived at a stock that had returned 15.3% over twelve months while its quarterly revenue doubled.

The de-rating created the room. Our three prior notes marked the forward multiple at 25.8x, then 23.5x, then 24.8x, so it ground sideways-to-lower across the coverage period while the estimate behind it rose 26%. That refusal to re-rate is exactly what made an 8.7% move possible on guidance rather than on a beat. The market was not paying for extrapolation, so when management supplied a committed number instead of an extrapolation, the gap closed violently.

What the tape did not price. The same prepared remarks contained a 350 basis point gross margin reset, a $27.2B sequential decline in free cash flow, days sales outstanding missing management's own forecast by five days, and $108.5B of guarantee exposure. A session that rises 8.7% on the revenue line while those four disclosures land in the same hour is pricing growth and deferring quality. We think the growth is real and we think the deferral is a choice the market will revisit, most likely at the Q4 margin trough.

Street Perspective

Debate: Is a Supply-Constrained 70% Guide Conservative or Committed?

Bull view: Management described demand consistent with roughly 100% growth and guided to what it can build, backed by $279B of supply commitments and a supply chain it has worked years upstream. Every incremental yield improvement and capacity addition is upside to a number that has no demand risk embedded in it.

Bear view: A company that has never guided a year out has just made its first attempt at the hardest forecast in the industry, across a product transition, against a memory market it admits surprised it once already. The guide converts what were quarterly judgements into a single annual commitment that can now be missed publicly.

Our take: The bull has the better argument and the bear identifies the real change. A supply-constrained guide genuinely does carry different risk than a demand forecast, and the commitment book substantiates it. But management has now staked its credibility on a number twelve months out, and the same memory market that forced this quarter's margin reset is a direct input to whether the supply materialises. We treat the ~70% as achievable and the reporting risk around it as newly concentrated.

Debate: Is Balance-Sheet-Supported Demand a Feature or a Warning?

Bull view: The frontier labs are constrained by compute rather than by customers, the platform is fungible and redeployable if any single lab fails, third-party institutional capital is doing the actual underwriting through the $500B financing platforms, and NVIDIA earns equity returns alongside hardware revenue. Nearly $50B invested is a small fraction of expected free cash flow over the same period.

Bear view: Roughly a quarter of next year's revenue depends on customers whose purchases lean on NVIDIA's balance sheet, guarantee exposure reaches $108.5B with $105B against a single counterparty's leases, and the company issued $25B of debt in a quarter it generated $21B of free cash flow. Vendor financing at this scale has ended badly in every prior technology cycle.

Our take: The fungibility argument is the crux and it is genuinely strong. If a lab fails, the racks redeploy into a market management describes as supply-constrained at least through the end of fiscal 2028, which is not how prior vendor-financing episodes ended. But we do not accept that this changes nothing. A quarter of revenue is a large exposure, the guarantee has no disclosed ceiling given the option on additional gigawatts, and management's own payback evidence this quarter was anecdotal. We are marking this as the thesis's primary risk going forward, displacing gross margin.

Debate: Does the Margin Reset Change the Quality of the Franchise?

Bull view: Falling from 75% to a 71% to 72% trough and recovering to 72% to 73% still leaves NVIDIA with among the highest gross margins in semiconductors, on revenue growing 70%. Memory scarcity is a symptom of the demand surge driving the growth, and executed price increases already restore most of it.

Bear view: A company that could not hold a framework it reaffirmed three quarters running has less pricing power over its own inputs than the platform narrative implies. Every 100 basis points is roughly $6.9B of annualised gross profit at the guided fiscal 2028 revenue base.

Our take: The bull is right on levels and the bear is right on what it reveals. The absolute margin remains extraordinary and the recovery has a named mechanism in executed price increases, which is stronger than a hope. What changed is that NVIDIA has demonstrated it will absorb input inflation for several quarters before passing it through, which is a real constraint on a business that had appeared to have none. We would treat a Q4 print below 70% as the signal that the pass-through is not working.

Model Update & Valuation Framework

Valuation. At the August 27 close of $227.98, NVDA trades at approximately 24.8x our fiscal 2027 non-GAAP EPS estimate of $9.20 and approximately 14.3x our fiscal 2028 estimate of $15.90. The fiscal 2028 figure is not an extrapolation: it is built on the revenue growth the company has formally guided and the gross margin range it has formally guided, which makes it the most substantiated forward estimate we have been able to construct on this name.

Fair value range: $285 to $350, raised from $250 to $300. The range reflects 18x to 22x our fiscal 2028 estimate of $15.90. The midpoint of $317 implies approximately 39% upside from $227.98. We have used a deliberately modest multiple range for a company guiding 70% growth, because the quality of the earnings has deteriorated in ways the multiple should reflect: cash conversion at 22.2%, receivables extending, and roughly a quarter of the guided revenue leaning on NVIDIA's own balance sheet.

What would change our view. Q4 gross margin printing below 70%. Days sales outstanding past 65 with receivables again outgrowing revenue, which would make part of the growth rate a credit decision rather than a sales one. Any expansion of guarantee exposure without a disclosed ceiling. Evidence that the Vera Rubin ramp is falling behind the ~20% of Q3 Data Center revenue guided. On the upside, a measured and sourced customer-payback disclosure, a fiscal 2028 volume-versus-price split, or Q4 margin holding above the guided trough.

Thesis Scorecard Post-Earnings

Overall: The growth case got materially stronger and the quality case got materially weaker, and neither cancels the other. A first-ever full-year guide at roughly 70%, backed by $279B of supply commitments, is the strongest forward disclosure this company has ever made. In the same hour it broke a margin framework it had reaffirmed three times, reported a $27.2B collapse in free cash flow conversion, missed its own receivables forecast, and disclosed $108.5B of guarantee exposure. Bull #4 moves to challenged and two new pillars enter the thesis, one on each side.

Action: Buy, with lower conviction than a quarter ago. On the guided fiscal 2028 revenue and the guided fiscal 2028 margin, non-GAAP earnings reach roughly $15.90 and the stock trades at 14.3x a year management has underwritten with $279B of purchase obligations. That is worth owning. But NVIDIA is becoming a capital provider to its own end market, and that carries a different risk profile than the platform-monopoly thesis we held three months ago. We maintain Outperform, raise the fair value range to $285 to $350 on a deliberately modest 18x to 22x multiple, and replace gross margin with balance-sheet-supported demand at the top of the watch list.