The Margin Fear Died and a Capital-Allocation Question Replaced It: $25.8 Billion of Strategic Outlays Against a $3.8 Billion Buyback

Results vs. Consensus

Q4 FY2026 Scorecard

Year-Over-Year Comparisons

Quarter-Over-Quarter Comparisons

Fiscal 2026 Full Year

The full-year gross margin of 71.3% is depressed by the $4.5B H20 inventory charge taken in Q1 FY26 and the $180M release in Q2. Excluding those, the underlying trajectory is the quarterly sequence of 72.7%, 73.6% and 75.2% across the last three quarters. The four reported quarters sum to the $215,938M full-year figure exactly.

Quality of Beat

Revenue. Two things about this print are more interesting than the beat itself. The first is that the company exceeded its own guide by $3.1B, which is nearly double its typical margin of conservatism, in a quarter where it had already told investors it was supply-constrained. The second is the mix: networking at $11.0B grew 263% year-over-year and now represents 17.6% of Data Center revenue, up from 16.0% a quarter ago and 8.5% a year ago. The business is becoming less of an accelerator company and more of a systems company at a measurable rate.

Margins. We flagged fiscal 2027 gross margin as the most important unpriced risk on the November call, on the strength of a verb: management had moved from saying it would deliver the mid-seventies to saying it would "work to try and hold" them. This quarter answered it. Non-GAAP gross margin printed 75.2%, twenty basis points above the 75.0% guide midpoint, Q1 is guided to 75.0%, and management stated it continues to see mid-seventies margins for the full year. We were right to flag it and wrong about the direction, and we would rather record that plainly than quietly drop the concern.

EPS. The cleanest way to read this quarter's earnings is through free cash flow, which grew 125% year-over-year to $34.9B and represented 51.2% of revenue. That is a company converting more than half of every revenue dollar into cash while growing that dollar 73%. What it did with the cash is a separate question, and it is the most important question this quarter raised.

Segment Performance

Revenue by Reportable Segment

Revenue by Market Platform

Data Center: $62.3B, 91.5% of Revenue

Data Center added $11.1B sequentially, its largest ever absolute increase, and finished fiscal 2026 at $193.7B, up 68%. Management put a scale marker on it: the business has grown roughly thirteen times since the emergence of ChatGPT in fiscal 2023. Grace Blackwell systems accounted for roughly two-thirds of Data Center revenue in the quarter, and nearly nine gigawatts of Blackwell infrastructure is now deployed and consumed by cloud service providers, hyperscalers, model makers and enterprises. Demand extends to prior generations: management said even Hopper and much of the six-year-old Ampere fleet are sold out in the cloud.

"With NVIDIA infrastructure in high demand, even Hopper and much of the 6-year-old Ampere based products are sold out in the cloud."
— Colette Kress, Executive Vice President and Chief Financial Officer

The single most consequential new disclosure was the customer mix. Management said the top five cloud providers and hyperscalers collectively account for slightly over 50% of Data Center revenue, and that growth in the quarter was led by everyone else: model makers, enterprises, supercomputing and sovereigns. Sovereign AI alone exceeded $30B in fiscal 2026 and more than tripled year-over-year, driven by customers in Canada, France, the Netherlands, Singapore and the United Kingdom.

Assessment: Disclosing hyperscaler concentration at the exact moment the market decided concentration was a reason to sell is either good timing or good governance, and either way it helps. The more durable point is that the non-hyperscaler half is growing faster, which converts our Bull #5 from an assertion into a measured fact.

Networking: $11.0B, Up 263% Year-Over-Year

Networking crossed $10B in a quarter for the first time and finished the year above $31B, which management framed against the year it acquired Mellanox: more than ten times the fiscal 2021 scale. Scale-up and scale-out both grew double digits sequentially, driven by NVLink, Spectrum-X Ethernet and InfiniBand. The year-over-year growth was led primarily by NVLink 72 scale-up switches, which follows mechanically from Grace Blackwell being two-thirds of Data Center revenue: each rack carries nine switch nodes with two chips each.

Two structural developments landed in the quarter. NVIDIA announced it will enable AWS to integrate NVLink with the hyperscaler's own custom silicon, and management claimed the company is now, or will shortly be, the largest Ethernet networking company in the world.

"The invention of NVLink really turbocharged our networking business. Every rack comes with 9 nodes of switches."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The AWS NVLink arrangement is the underrated item. Licensing the interconnect to a customer building its own accelerator converts a competitive threat into an attach opportunity, and it means NVIDIA earns content even in racks where it loses the compute socket. That is a materially better answer to the custom-silicon question than any performance benchmark.

Gaming: $3.7B, Down 13% Sequentially, and Now Supply-Rationed

Gaming grew 47% year-over-year but fell 13% sequentially as channel inventory moderated after a strong holiday season. Management then said something more important than the quarter: it expects supply constraints to be a headwind to Gaming in Q1 fiscal 2027 and beyond, while describing end demand as strong and channel inventory as healthy.

Asked directly whether Gaming can still grow year-over-year in fiscal 2027, management declined to commit, saying supply will be very tight for a couple of quarters and that an improvement late in the year could open the possibility.

Assessment: This confirms the supply-allocation hypothesis we raised last quarter. Gaming is not weakening, it is being rationed, and the rationing is a rational choice when Data Center gross margins are 75% and the same memory and packaging capacity feeds both. Investors should model Gaming as a residual claimant on capacity rather than as a demand-driven line.

Professional Visualization and Automotive: $1.9B Combined

Professional Visualization crossed $1B for the first time, at $1,321M, up 159% year-over-year and 74% sequentially on exceptional Blackwell demand and the RTX PRO 5000 Blackwell workstation launch. Automotive grew 6% year-over-year to $604M, a marked deceleration from the 32% growth of the prior quarter. Management separately disclosed that physical AI contributed north of $6B of revenue in fiscal 2026, a figure that spans segments rather than sitting inside Automotive.

Assessment: The Professional Visualization inflection is real but small, and it is better read as evidence that Blackwell demand exceeds Data Center. Automotive's deceleration to 6% deserves a note: the segment has an attractive long-dated story in robotaxi and Alpamayo, and a near-term growth rate that no longer supports it.

Key Topics & Management Commentary

Overall Management Tone: Confident and expansive on demand, and conspicuously unforthcoming on capital allocation. The prepared remarks resolved last quarter's margin question directly and without hedging, and volunteered a first-ever hyperscaler concentration figure, both of which are the marks of a management team that feels it is winning the argument. On the two questions where investors most needed specificity, the size and rationale of the ecosystem investment programme and the terms of a $13.0B transaction that appears as its own line in the cash flow statement, the answers were thematic and the numbers were absent.

1. The Margin Question From Last Quarter Is Answered

Non-GAAP gross margin came in at 75.2% against a 75.0% guide midpoint, the third consecutive quarter of sequential expansion, and Q1 is guided to 75.0%. Management went further and reaffirmed the framework for the whole of fiscal 2027, adding only that it would update investors as the Vera Rubin transition approaches.

Asked how sustainable the mid-seventies are over a longer horizon, the answer reframed margin as an output of the product cadence rather than an input to be defended.

"The single most important lever of our gross margins is actually delivering generational leads to our customers."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: This is the right frame and it is also the honest one. If each generation delivers multiples of performance per watt and per dollar, the customer's cost per token falls even as NVIDIA's price holds, and the margin is defensible without a pricing fight. The framework fails only if a generation disappoints, which makes the Vera Rubin ramp the thing to watch rather than the memory market.

2. The $13.0 Billion Groq Line, Explained on the Call and Nowhere Else

The cash flow statement carries a separately captioned line in investing activities reading "Groq, Inc." at $13,000M for the quarter. The transaction itself is named in the release's Data Center highlights, as a non-exclusive licensing agreement with Groq to accelerate AI inference at global scale, and the CFO commentary attributes part of the quarter's cash movement to outlays for an intellectual property license. Neither names a figure. The $13,000M appears only in the cash flow statement, and the only narrative sizing of it anywhere is the closing paragraph of the CEO's remarks.

"Finally, we recently entered into a nonexclusive licensing agreement with Grok for its low latency inference technology and welcome the team of brilliant engineers to NVIDIA. As we did with Mellanox, we will extend NVIDIA's architecture with Grok's innovations to enable new levels of AI infrastructure performance and value."
— Jensen Huang, Founder and Chief Executive Officer

The counterparty is Groq, Inc., as captioned in the filed cash flow statement; the call transcript renders the name phonetically. The structure is a non-exclusive technology licence plus the engineering team rather than an acquisition, and management deferred all detail to its March GTC event. Asked later how a decode-specific solution fits the roadmap, the answer was again deferred, with the Mellanox analogy repeated.

Assessment: Two readings, and both are worth holding. The generous one is that NVIDIA bought the best low-latency decode team in the industry for a little over a third of a quarter's free cash flow and will fold it into CUDA as it did with Mellanox, which worked spectacularly. The uncomfortable one is that NVIDIA paid $13.0B to license inference technology from a company whose entire thesis was that general-purpose GPUs are the wrong economic tool for inference. Paying that much is an acknowledgement that the workload class is real. The disclosure standard here is the genuine problem: a sum equal to 37% of quarterly free cash flow, sized in one cash flow line and explained in one sentence of prepared remarks.

3. Capital Allocation Quietly Changed

NVIDIA repurchased $3,815M of stock in the quarter, down from $12,456M in Q3. In the same quarter it spent $12,800M on non-marketable equity securities, up from $478M in the year-ago quarter, and $13,000M on the Groq transaction. The full-year figures tell the same story: $17,502M of non-marketable equity purchases in fiscal 2026 against $1,486M in fiscal 2025.

Asked why a company generating roughly $100B of annual cash and trading at a price the questioner characterized as attractive would not commit to a much larger repurchase, management defended the investment priority without ranking it against buybacks.

Q: "So I was wondering if you can talk about that like question being, why not put a big stake in the ground and just have a huge share repo here?"
— Timothy Arcuri, UBS

A: "So we will continue to make this a very important part of our process and strategic investments. But of course, we are still repurchasing our stock."
— Colette Kress, Executive Vice President and Chief Financial Officer

For the year the company returned $41.1B, which management characterized as 43% of free cash flow, comprising $40.1B of repurchases and $974M of dividends. The great majority of that repurchase happened in the first three quarters.

Assessment: This is the quarter's real news and it was not presented as news. A company that returned $12.7B to shareholders in Q3 returned $4.1B in Q4 while committing $25.8B to strategic outlays. We are not against the strategy: the ecosystem logic is coherent and NVIDIA's cost of capital is low. We are against learning about a shift of this size by differencing two cash flow statements.

4. Stock-Based Compensation Enters Non-GAAP

Beginning in Q1 fiscal 2027, NVIDIA will include stock-based compensation in its non-GAAP measures, describing it as a foundational component of its compensation programme. The Q1 guide is presented on the new basis: non-GAAP operating expenses of approximately $7.5B inclusive of $1.9B of stock-based compensation, and a non-GAAP gross margin of 75.0% inclusive of a 10 basis point impact.

The change was welcomed on the call before the questioner moved to their actual topic.

"First, let me say kudos on including the stock comp in non-GAAP. I think that's a great move."
— Benjamin Reitzes, Melius Research

Assessment: Unambiguously good governance and a modest analytical nuisance. It also creates a trap in the fiscal 2027 numbers: Q1 non-GAAP operating expenses of $7.5B against Q4's $5,102M looks like 47% sequential growth, when the like-for-like figure excluding the $1.9B of stock-based compensation is roughly $5.6B, or about 10%. Anyone comparing the guided line to last year's without adjusting will conclude the cost base has exploded. It has not.

5. The $500 Billion Framework Is Now Explicitly Exceeded

The cumulative Blackwell and Rubin revenue framework introduced last year has been overtaken. Management said it expects sequential revenue growth throughout calendar 2026 at a level exceeding what the $500B opportunity contemplated, and that inventory and supply commitments are in place for demand extending into calendar 2027.

"We look ahead, we expect sequential revenue growth throughout calendar 2026, exceeding what was included in the $500 billion Blackwell and Rubin revenue opportunity we shared last year."
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: We wrote last quarter that the framework should be treated as a floor. It was. The commitment to sequential growth in every quarter of calendar 2026 is a stronger statement than the framework itself, because it is falsifiable within 90 days rather than over eight quarters.

6. Rubin Ships Samples, Production in the Second Half

The Rubin platform was unveiled at CES: six new chips comprising the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet switch. Management said the platform will train mixture-of-experts models with a quarter of the GPUs and reduce inference token costs by up to ten times against Blackwell. First Vera Rubin samples shipped to customers in the week of the call, with production shipments on track for the second half of the year, and AWS, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure among the first to deploy.

Asked whether the Rubin ramp will produce a sequential-growth step change comparable to Blackwell's, management declined to size it, saying it is too early to determine how much of the second half the ramp represents while stating that essentially every customer is expected to purchase Vera Rubin.

Assessment: A modular, cable-free tray design that improves serviceability while remaining architecturally compatible is the practical expression of the annual cadence. The unanswered question is the shape of the transition, and it matters: a Blackwell-like ramp would put fiscal 2027 revenue materially above our estimate, while a slower stand-up would concentrate growth in fiscal 2028.

7. Agentic AI Named as the Inflection, With Named Evidence

Management's demand argument has moved from capability to economics. It cited Claude Code, Claude Cowork and OpenAI Codex as having achieved useful intelligence, described adoption as skyrocketing, and, crucially, said the tokens are profitable. On the return-on-investment question it again pointed to Meta, whose GEM model advancements drove a 3.5% increase in ad clicks on Facebook and more than a 1% gain in conversions on Instagram.

"In this new world of AI, compute is revenues. Without compute, there's no way to generate tokens. Without tokens, there's no way to grow revenues."
— Jensen Huang, Founder and Chief Executive Officer

Analyst expectations for 2026 capital expenditure across the top five cloud providers and hyperscalers are now approaching $700B, up nearly $120B since the start of the year.

Assessment: The token-profitability claim is the load-bearing one and it remains asserted rather than evidenced at the customer level. The Meta figures are measured and useful, and they are also the same single example cited last quarter with different metrics. One named customer proof point across two consecutive calls is thinner than the confidence of the argument implies.

8. Supply Commitments Nearly Double Again

Total supply-related commitments rose to $95.2B from $50.3B, an 89% sequential increase on top of the 63% increase the prior quarter. Inventory rose 8% sequentially to $21.4B, a marked deceleration from Q3's 32%. Management framed the commitment build as deliberate and unusually long-dated.

"We have strategically secured inventory and capacity to meet demand beyond the next several quarters. This is further out in time than usual and reflects the longer demand visibility we have."
— Colette Kress, Executive Vice President and Chief Financial Officer

Days sales outstanding improved again, to 51 days from 53, so the working-capital build remains supply-side rather than receivables-driven.

Assessment: Commitments have grown from roughly $31B to $95.2B in two quarters while quarterly revenue grew from $46.7B to $68.1B. The commitment book is now compounding faster than revenue, which is the correct posture if the demand signal holds and an expensive one if it does not. The improving DSO is the reassuring counterweight and it is worth more than the commitment number is worth worrying about.

9. China Moves From Zero to a Different Kind of Zero

Small quantities of H200 were approved for China-based customers by the United States government, and NVIDIA has generated no revenue from them. Management said it does not know whether any imports will be permitted. The Q1 guide again assumes no Data Center compute revenue from China.

The more significant commentary was competitive rather than regulatory: management said Chinese competitors, bolstered by recent public offerings, are making progress and have the potential to disrupt the structure of the global AI industry over the long term.

Assessment: This is a meaningful escalation in candour from the prior quarter's framing. An export approval that produces no revenue because the destination market may not accept the product is a different situation from an export ban, and it is worse: it suggests the option value we have been carrying at zero may be closer to zero permanently. We continue to model China at nil and now treat a reopening as a low-probability event rather than a deferred one.

10. Capital Returns and the Fiscal 2026 Scorecard

For the full year NVIDIA generated $96.6B of free cash flow and returned $41.1B, or about 43%, through $40.1B of repurchases and $974M of dividends. It ended the year with $62.6B of cash and marketable securities and $58.5B remaining under its repurchase authorization, down from $62.2B a quarter earlier.

Assessment: A 43% payout ratio on free cash flow is respectable in isolation and misleading in trend. The rate through the first three quarters was far higher, and the fourth quarter's $4.1B against $34.9B of free cash flow is a 12% payout. The authorization barely moved because the buyback barely ran.

Guidance & Outlook

The $78.0B guide is the largest single forward step the company has ever taken, adding $9.9B of sequential revenue against the $11.1B it just delivered, and it sits 7.4% above where the Street had modelled the quarter. Against Q1 fiscal 2026's $44,062M it implies 77% year-over-year growth. Management paired it with a commitment to sequential growth in every quarter of calendar 2026 and with the reaffirmed mid-seventies gross margin framework.

Implied Q-over-Q ramp: Sequential growth of 14.5% is a modest deceleration from the 19.5% just delivered, on a base 20% larger. If the sequential rate decays through the year toward the high single digits as Vera Rubin transitions in, fiscal 2027 revenue lands near $350B, or roughly 62% growth.

Street at: Consensus for Q1 was $72.6B, so the guide sits $5.4B above it. The more important gap is what the guide implies for the full year against a Street that had not yet marked to the sequential-growth commitment.

Guidance style: The pattern is now clearly established and it changed this quarter. NVIDIA has historically beaten its guide by low single digits; this quarter it beat by 4.8%, and it exceeded even the top of its own guided band. Combined with the willingness to commit to sequential growth across four unguided quarters, this is a management team that has concluded its constraint is supply and is signalling that the supply is now secured. The $95.2B commitment book is the collateral for that signal.

Analyst Q&A Highlights

Whether Customers Can Keep Funding Capital Expenditure at This Level

The opening question went straight to the bear case: hyperscaler capital expenditure approaching $700B is hard to grow from, several large customers face compressing cash generation, and the question was whether NVIDIA can still grow if that envelope stops expanding.

Q: "how confident are you about your customers' ability to continue to grow their CapEx? And if their CapEx doesn't grow can NVIDIA still find a way to grow in that envelope?"
— Vivek Arya, Bank of America Securities

A: "I am confident in their cash flow growing."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: Management answered the first half with conviction and skipped the second half entirely. The argument advanced was that agentic AI has inflected, compute translates directly into tokens and tokens into revenue, so customer cash flows grow rather than compress. That is a coherent case. It is also not an answer to what happens if it is wrong, and the questioner explicitly asked for that scenario. The absence of a downside plan is a recurring feature of these calls.

The Role of the Balance Sheet in Ecosystem Investment

A question on the strategic investment programme asked management to explain the role of stakes taken across model developers, cloud partners and silicon and software companies, and how it views the balance sheet as an instrument for extending its ecosystem position.

Q: "Can you talk about the role of those investments? And kind of how do you view the balance sheet as a tool to kind of grow the NVIDIA's position in the ecosystem and participate in that growth?"
— Joseph Moore, Morgan Stanley

A: "So our investments are focused very squarely, strategically on expanding and deepening our ecosystem reach."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The answer described the philosophy at length and disclosed nothing measurable. No aggregate size, no ceiling, no return threshold, no indication of what share of revenue comes from entities NVIDIA holds a stake in. This is the second consecutive quarter in which the question has been asked plainly and answered thematically, and the amounts have grown roughly threefold in between. At $12.8B of equity purchases in a single quarter the programme is no longer small enough for a thematic answer to suffice.

Whether the Roadmap Is Moving Toward Workload-Specific Silicon

The most probing question of the call connected two facts management had disclosed separately: the introduction of a long-context-specific product and the acquisition of a low-latency decode technology licence. The implication was that NVIDIA is itself moving toward the workload-specialised silicon it argues against.

Q: "Curious how we should think about your future road map? Should we be thinking about customized silicon either by workload or customer as an increasing focus by NVIDIA"
— Christopher Muse, Cantor Fitzgerald

A: "I've got some great ideas that I'd like to share with you at GTC. But the simple idea is that our infrastructure is incredibly versatile because of CUDA, and we're going to continue to do that."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: A deferral, and the most consequential one on the call. The architectural-compatibility argument that followed is genuinely strong: because every GPU generation shares an architecture, software optimisation done for the newest part also lifts the installed base, which is why six-year-old silicon still commands rent. But that argument explains why NVIDIA does not need workload-specific parts, and the company just spent $13.0B acquiring the capability to build one. The question was correct and it remains open until March.

The Shape of Sequential Growth and Whether Gaming Grows Next Year

A two-part question asked how the sequential-growth commitment plays out across the year as Vera Rubin ramps into the second half, and whether Gaming can grow year-over-year in fiscal 2027 given the memory situation.

Q: "Do you think gaming can still grow year-over-year in fiscal '27? Or will that be under more pressure given memory?"
— Stacy Rasgon, Bernstein Research

A: "As much as we would love to have additional more supply, we do believe for a couple of quarters, it is going to be very tight."
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: The Gaming half was answered honestly and unhelpfully, which is the right combination. The Rubin half was not answered: management said it is too early to size the second-half ramp while confirming that essentially every customer intends to buy. For modelling purposes that leaves the fiscal 2027 second half genuinely uncertain in shape even as the direction is not in doubt.

Whether CUDA Still Matters as Spending Shifts to Inference

A question tested the durability of the software moat against the argument that inference is a more commoditisable workload than training, and therefore less dependent on the CUDA ecosystem.

Q: "Jensen, I'm curious if you can touch on the importance of CUDA as now more of the investment dollars in AI are coming from inference workloads."
— Atif Malik, Citi

A: "Without CUDA, we wouldn't know what to do with inference."
— Jensen Huang, Founder and Chief Executive Officer

Assessment: The substance behind the soundbite is the specific claim that NVLink 72's 50x performance-per-watt gain required inventing new parallelisation algorithms that sit on top of CUDA to distribute inference across aggregate bandwidth. That is a concrete, checkable mechanism rather than a general moat assertion, and it is the strongest version of the argument. It also sits awkwardly beside a $13.0B licence for someone else's inference technology.

Revenue Diversification and the Hyperscaler Share

A question sought clarification on the new concentration disclosure and what it implies about the trajectory of the customer base.

Q: "does that imply your non-hyperscale customers grew faster? And if so, what are the -- can you help us understand what are the non-hyperscalers doing different?"
— Mark Lipacis, Evercore ISI

A: "There's a big organization, therefore, of diversity of all different other types of companies that we are working with, that it goes through our AI model makers, that goes through our enterprises, that goes to supercomputing, it goes to our sovereigns."
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: Confirmed, and this is the most useful exchange of the call for the bear case. The non-hyperscaler cohort is growing faster than the hyperscaler cohort, which directly addresses the concentration criticism that drove the stock lower the following day. One caution on precision: the Q&A answer described the top five as about 50% of total revenue while the written CFO commentary specifies slightly over 50% of Data Center revenue. The written figure is the one to model.

Why Not a Much Larger Buyback at This Price

The final substantive challenge noted that the company will generate roughly $100B of cash this year, that the purchase commitments appear to be past their peak build, and that the stock has not appreciated despite consistently strong results, and asked why that does not argue for a far larger repurchase.

Q: "So I was wondering if you can talk about that like question being, why not put a big stake in the ground and just have a huge share repo here?"
— Timothy Arcuri, UBS

A: "We look at our capital return very, very carefully, and we do believe that one of the most important things that we can do is really supporting the extreme ecosystem that's in front of us"
— Colette Kress, Executive Vice President and Chief Financial Officer

Assessment: The question embedded the observation that the market is not rewarding results, and management did not engage with that framing at all. The answer ranked ecosystem support and supplier capacity above repurchase without saying so explicitly, which is a defensible allocation and an evasive presentation. Given that the quarter's buyback fell 69% sequentially while strategic outlays reached $25.8B, investors were entitled to hear the priority stated rather than inferred.

What They're NOT Saying

1. The terms, rationale or accounting for the $13.0B Groq transaction: One sentence of prepared remarks, one captioned but unexplained cash flow line, and a deferral to GTC. No disclosure of what portion is licence versus compensation, over what period it will be amortised, or what it does to the roadmap.
2. Any size or ceiling on the ecosystem investment programme: Asked twice, answered thematically both times. Non-marketable equity purchases were $12.8B in the quarter and $17.5B for the year against $1.5B in fiscal 2025, and management still will not frame the programme in dollars.
3. What share of revenue comes from entities NVIDIA has invested in: The company disclosed hyperscaler concentration for the first time this quarter, which makes the continued absence of this figure a choice rather than an oversight.
4. The shape of the Vera Rubin ramp: Management confirmed second-half production and universal customer intent, and explicitly declined to size the second-half contribution. That leaves the largest swing factor in fiscal 2027 unquantified.
5. Whether Gaming grows in fiscal 2027: Asked directly, answered with a supply description and a deferral. Given that Gaming is now explicitly supply-rationed, the honest answer is that it depends on an allocation decision the company has already made in Data Center's favour.
6. What happens if hyperscaler capital expenditure stops growing: The question was asked in two parts and only the optimistic part was answered. There is no articulated scenario in which NVIDIA grows into a flat customer capital expenditure envelope.
7. The fiscal 2027 operating expense base on a comparable footing: Management guided full-year non-GAAP operating expense growth in the low forties on the new stock-based-compensation-inclusive basis without restating the prior-year comparison, which leaves the growth rate uninterpretable without reconstructing the base.

Market Reaction

• Pre-print setup: NVDA closed at $195.56 on February 25, up 4.9% year-to-date, up 4.9% over the trailing thirty days, and up 54.4% over the trailing twelve months. The stock entered the print roughly 6% below its 52-week closing high of $207.04.
• After-hours move: Shares initially rose in extended trading on the print, then gave back most of the gain before the session ended.
• Next-day session: The stock opened at $194.27, down 0.7%, traded in a range of $184.32 to $194.29, and closed at $184.89, down 5.5% or $10.67 from the pre-print close. The close was at the low end of the day's range.
• Volume: 360.8 million shares against a 30-day average of 171.6 million, a 2.1x multiple and heavier than the 1.8x that accompanied last quarter's print.
• Market and peer context: The S&P 500 fell 0.5% and the Nasdaq Composite fell 1.2% in the same session. AMD fell 3.4% and Intel fell 3.0%.

This is the second consecutive quarter in which NVDA beat on every line, raised the forward bar well above the Street, and fell. It is also materially worse than the first: in November the S&P fell 1.6% alongside it and roughly half the move was beta, whereas here the index fell 0.5% and the stock fell 5.5%. Whatever this session was pricing, it was specific to NVIDIA and to the AI complex rather than to the market.

What the tape was pricing. Three concerns circulated and none of them concerned the quarter. The first was the emergence of a credible alternative supplier, with Google supplying Anthropic with its own accelerators and reported to be in discussions with Meta. The second was customer concentration, which is a curious thing to sell on in the exact quarter management disclosed the figure and showed the non-hyperscaler cohort growing faster. The third was the durability of infrastructure spending generally.

Our reading. The Anthropic point is the substantive one and it deserves to be held precisely. Anthropic committing to NVIDIA infrastructure and taking capacity from an alternative supplier are not contradictory: a frontier lab at this growth rate buys everything available. But it does establish that a large customer will run frontier workloads on non-CUDA silicon, which is the first concrete crack in an argument NVIDIA has been making unopposed. That is a legitimate reason to demand a lower multiple. It is not a reason to sell a company that just guided to 77% growth at 23 times forward earnings.

Street Perspective

Debate: Does an Alternative Accelerator Winning a Frontier Lab Break the Moat?

Bull view: Frontier labs are capacity-constrained and buy every viable accelerator they can secure, so a second supplier is additive rather than substitutive. NVIDIA remains the only architecture that runs every model in every cloud, and it just licensed the leading low-latency decode technology to close the one gap that existed.

Bear view: The relevant fact is not volume but proof. A frontier lab running production workloads on a rival architecture demonstrates that the CUDA switching cost is finite and payable, and once it is paid the second deployment is cheaper than the first.

Our take: The bear has identified a real change and is overstating its near-term consequence. The switching cost is finite, but the offtake argument still holds: a lab can move workloads to a captive supplier's silicon, while a neocloud building merchant capacity cannot, because it needs a rentable demand pool. We expect share erosion at the frontier and continued dominance in the merchant market, which is the larger pool.

Debate: Is the Shift From Buybacks to Ecosystem Investment Good Capital Allocation?

Bull view: NVIDIA is buying pre-inflection equity in the companies defining the compute era, at prices that will look trivial in retrospect, and locking in architectural commitments in the process. Buying back stock at 23 times forward earnings is a worse use of a dollar than owning a share of the demand.

Bear view: A $25.8B quarterly outlay disclosed through cash flow line items, against a buyback cut by more than two-thirds, is a governance failure regardless of whether the investments are wise. Investors cannot underwrite a strategy that management will not size.

Our take: We think the strategy is probably right and the disclosure is definitely wrong, and only one of those is fixable by management in the next 90 days. The bear's argument is not about the merits and does not need to be. We are treating this as an escalated risk rather than a thesis break, and we will treat a third quarter of the same opacity as something more serious.

Debate: Can Gross Margin Hold the Mid-Seventies Through the Vera Rubin Transition?

Bull view: Management has now delivered three consecutive quarters of sequential expansion, printed above its own guide midpoint, and reaffirmed the framework for the full year with the transition in view. The margin lever is generational performance rather than pricing, and Rubin's stated ten times reduction in inference token cost is exactly that lever.

Bear view: Every rack-scale transition has compressed margin on the way in, Blackwell's did so by more than 400 basis points at the trough, and Rubin introduces six new chips at once against a memory market that is tightening.

Our take: We were on the bear side of this debate last quarter and the company proved us wrong, so we weight its guidance more heavily now. The Blackwell trough is the right base rate and the mitigating difference is real: Rubin is architecturally compatible and the supply chain has already learned the rack. We model a shallower dip than Blackwell's and we would treat any Q2 or Q3 guide below 73% as the signal that the base rate won.

Model Update & Valuation Framework

Basis note. Our fiscal 2027 estimate is struck on the new non-GAAP definition that includes stock-based compensation, matching how the company will report from Q1. Comparing it to the fiscal 2026 figure of $4.77, which excludes stock-based compensation, understates the growth by roughly six percentage points. On a like-for-like basis our fiscal 2027 estimate is approximately $8.15.

Valuation. At the February 26 close of $184.89, NVDA trades at approximately 38.8x fiscal 2026 non-GAAP EPS of $4.77 and approximately 23.5x our fiscal 2027 estimate of $7.88. The forward multiple has compressed from 25.8x at our initiation three months ago even though the share price rose 2.4% over the same period, because our estimate rose 12.6% and the price did not keep pace. That is the mechanical definition of a stock getting cheaper while the business improves, and the reaction-day declines that produced it have now happened twice in a row.

Fair value range: $220 to $270, raised from $210 to $250. The range reflects 28x to 34x our fiscal 2027 estimate of $7.88. The midpoint of $245 implies approximately 33% upside from $184.89. We have lowered the multiple range by two turns relative to our initiation's 30x to 36x because the earnings basis is now stricter, and raised the absolute range because the estimate moved more than the multiple came down.

What would change our view. A gross margin guide below 73% for any fiscal 2027 quarter, evidence that the Vera Rubin second-half ramp is slipping, a third consecutive quarter in which ecosystem investment commitments grow without disclosure, or a second frontier lab moving material production workloads to a rival architecture. On the upside, a GTC disclosure that frames the Groq technology as a rack-level attach rather than a defensive purchase, or a formal fiscal 2027 revenue framework, would move us toward the top of the range.

Thesis Scorecard Post-Earnings

Overall: Thesis strengthened on the numbers and complicated on the qualitative side. The single bear point we ranked highest at initiation, fiscal 2027 gross margin, has been resolved in the company's favour. In its place, two risks escalated that are both about what management chooses to disclose rather than what the business is doing: the unsized ecosystem programme and the unexplained Groq transaction. That is a better set of problems to have, and it is a worse set of problems to be given no information about.

Action: Buy. Two consecutive quarters of beating, raising, and falling have compressed the forward multiple to roughly 23x on a business the company has just committed to growing sequentially in every quarter of calendar 2026. We maintain Outperform, raise the fair value range to $220 to $270, and elevate capital-allocation disclosure to the top of the watch list for the May print. We would add on weakness toward the $170 area and would revisit the rating if a third quarter passes with $25B-scale outlays and no disclosure framework.