Nvidia's First Report Without Huang: $115.8 Billion in a Quarter, and a Margin Crack Is the Only Flaw
On 17 November 2026 Nvidia reported its first results since Jensen Huang resigned: quarterly revenue of $115.8 billion, more than double a year earlier, with data centers at 93% of sales and next-quarter guidance of $130 billion. But margin guidance fell to 71%–72%, and the stock rose only about 3% after hours.

The short version: a flawless quarter, and the market rose only 3%
After the US market closed on 17 November 2026, Nvidia reported results for the third quarter of fiscal 2027: revenue of $115.8 billion, up 103% year on year and 20% quarter on quarter; net income of $65.3 billion, up 105%; and adjusted earnings per share of $2.68, above the $2.48 the market expected. Data center revenue was $107.6 billion, up 110% year on year and 93% of total revenue. The company also guided next quarter's revenue to $130 billion, plus or minus 3%, above the $125.1 billion consensus, and, according to CNBC's account of the earnings call, raised its fiscal 2028 revenue growth outlook from 70% to 75%. It was the first earnings report since founder Jensen Huang resigned on 12 November, delivered by interim CEO Colette Kress, who remains chief financial officer.
The stock rose only about 3% in after-hours trading. That muted reaction is worth pausing over: almost every headline number beat expectations, and guidance beat too, yet the market did not surge. The reason is not this quarter but the next one — the company expects fourth-quarter gross margin to fall to 71% to 72%, about three points below this quarter's 74.8%, and the $130 billion guide contains an extra week, so on a normal 13-week basis sequential growth is only about 4%, not the 12% the headline implies. Before the report, Nvidia's shares had fallen about 24% in the month, down six weeks in a row. What the market really wanted to know was not "did this quarter sell well?" but "can Nvidia without Huang keep telling the growth story?" The report's answer: sales are excellent, but the next chapter comes with a discount.
The numbers: records on almost every line
The headline figures deserve to be stated plainly, because they are extraordinary. Third-quarter revenue was $115.8 billion, against $57.0 billion a year earlier and $96.2 billion last quarter. In other words, Nvidia's quarterly revenue more than doubled in a year and rose by nearly $20 billion in three months. Gross margin was 74.8%, up 1.4 points from 73.4% a year ago and down 0.2 points from 75.0% last quarter. Operating income was $77.5 billion, up 115% year on year; net income was $65.3 billion, up 105%, a record. Diluted earnings per share on a GAAP basis were $2.70, up 108%.
Those numbers would be explosive for any company; for Nvidia, the market has almost grown used to them — the company has set new records nearly every quarter for two years. But the context this time was different: the quarter ended on 25 October, when Huang was still in charge; by the time the results were published, he had been gone for five days. Investors reading this report were really asking two questions: is demand still there, and will execution suffer under new leadership? The first answer is written in the data center's 110% growth; the second, at least this quarter, shows no cracks at all.
The data center is the whole story
Nvidia's revenue structure can now be summed up in one sentence: the data center is the company. Data center revenue of $107.6 billion is 93% of the total; the remaining edge computing business — gaming graphics, PCs and robotics chips — brought in $8.2 billion, up 42% year on year, respectable but negligible next to the data center. Who buys from the data center division? The release does not name customers one by one, but on the earnings call the company said hyperscale cloud providers remain the largest group, while AI cloud, industrial and enterprise customers grew fastest, accounting for more than half of the data center's year-on-year increase. The buyers of Nvidia chips are no longer just a handful of cloud giants; they are a growing set of AI companies, industrial firms and frontier labs building their own data centers.
That shift in the customer base is good news and a risk at the same time. The good news is that demand is broadening beyond a few cloud providers' capital-spending cycles. The risk is that concentration remains high — according to CNBC's account of the call, four direct customers each accounted for 11% to 13% of revenue over the past three quarters. When a handful of buyers determine most of your revenue, their spending plans become your forecast. That is why Nvidia's share price moves every time a cloud company announces capital spending: the market is not watching how many chips Nvidia sold, but how much its customers plan to spend next year. There is a longer-term variable too: the release says Nvidia is expanding its collaboration with Meta on the Muse AI agent and its family of models, with Meta planning expanded deployment of Vera Rubin, Vera CPUs and networking gear to scale training, inference and agents. Big customers are both a source of revenue and co-authors of the technology roadmap — a depth of entanglement other chip companies can hardly replicate, but one that would also hurt more if a major customer shifted to its own chips.
The most important product this quarter is Vera Rubin NVL72. The release says the new rack-scale system has begun shipping in the fastest product ramp in the company's history, with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius, and coming online soon at SpaceX's infrastructure arm. Zacks reported that Vera Rubin generated nearly $20 billion in its debut quarter — a third-party figure not disclosed in the release itself, but if accurate it means the new platform is ramping faster than Blackwell did at the same stage. Whether Vera Rubin succeeds will decide whether Nvidia can keep growing at triple-digit rates for the next two years.
Gross margin: the only crack in the report
If this report has a soft spot, it is gross margin. The third quarter's 74.8% was down a fractional 0.2 points sequentially — not a problem in itself. The problem is next quarter's guide: 71% to 72%, a drop of roughly three points. The company attributed it to ramping production of Vera Rubin — new products cost more to build at first, before factories hit full speed and yields improve, so margins get dragged down. That is manufacturing common sense: any new chip is more expensive in early production, and margins recover as yields rise.
But the market only half-believed the explanation. The modest 3% after-hours gain is largely explained by the margin guide. Investors are really worried about two things: first, whether the margin decline will persist — if Vera Rubin's cost structure is inherently higher than the previous generation's, margins near 75% may never return; second, whether Nvidia's pricing power is loosening — as more customers develop their own chips and rivals' alternatives mature, can Nvidia still hold margins near 75%? The report answered neither question. Only the next quarter can.
A falling margin percentage, however, does not mean falling profit in dollars. At a 74.8% margin, third-quarter gross profit was about $86.6 billion; at the guide's midpoint of 71.5% on $130 billion of revenue, it would be about $93 billion. A lower percentage, more money. That is the power of scale: as long as revenue keeps growing fast, a modest margin retreat does no fundamental damage. What would be genuinely alarming is slowing revenue growth combined with falling margins — a combination that has not appeared yet. One more detail deserves attention: third-quarter operating expenses were $9.2 billion, up 57% year on year. The growth rate is below revenue growth, but the absolute figure is already large. Nvidia is investing simultaneously in next-generation platforms, robotics, autonomous driving and scientific computing; those investments lift costs now and are the price of staying ahead. Investors must judge whether they turn into revenue, not just the expense figure itself.
The $130 billion guide contains an extra week
The fourth-quarter revenue guide of $130 billion looks 12% higher than the third quarter, but the number hides a calendar quirk: the fourth quarter contains an additional week. Nvidia's fiscal calendar counts weeks; most quarters are 13 weeks, this one is 14. Scaled back to a normal 13 weeks, $130 billion becomes roughly $120.7 billion, sequential growth of only about 4%. That is analysts' rough arithmetic, not company guidance, and real sales do not spread evenly by week. But it is a reminder: the better a headline number looks, the more its basis matters. What investors should really watch is growth on an equal-week basis, and whether margin holds inside the 71% to 72% range.
The guide contains another notable assumption: the company stated explicitly that its outlook "is not assuming any Data Center compute revenue from China". The background is that the US Commerce Department moved the day before the report to restrict Chinese companies from renting compute, and Nvidia had already stopped selling licensed H200 chips to China. In other words, Nvidia has removed China from its growth model entirely — the $130 billion guide is built on an assumption of zero Chinese revenue. That makes the guide look conservative, but it also reflects a reality: geopolitics is now a fixed deduction in Nvidia's accounts, not a one-off surprise.
The first report without Huang
The other story in this report has nothing to do with numbers: it is the first Nvidia earnings release without Jensen Huang. Huang resigned as president, CEO and board member on 12 November, and the board accepted his resignation; new board chair Harvey Jones said at the start of the call that the board unanimously accepted the resignation and "does not intend to reinstate Jensen as CEO". Earlier the same day, Huang had said he would be "honoured" to return if the board asked. The board closed that door on the spot.
Kress set the tone on the call as "business as usual", saying customers, suppliers and partners are "100% behind us". Longtime executive Ian Buck joined the call to answer questions; some analysts see him as a CEO candidate. But the very phrase "business as usual" reveals what the market is worried about: Nvidia's story for thirty years has largely been Huang's personal story — his technical judgement, his grip on customers, his willingness to bet big at decisive moments. Kress, CFO since 2013, understands numbers and Wall Street, but whether she can replicate that founder-style leadership is unknown. The board's search for a permanent CEO is under way; until it concludes, every earnings report will be read through the lens of the post-Huang era.
One detail in the report relates directly to the transition: the company returned $28.1 billion to shareholders in the quarter through buybacks and dividends, with about $227.3 billion remaining under its repurchase authorisation, including a $150 billion increase the board approved in September. It also declared a quarterly dividend of $0.25 per share, payable on 23 December. Large buybacks and dividends are one way a company signals confidence — we have the cash, and we believe in our future. But an authorisation is permission, not a promise: the company may buy, or it may not. Real confidence will have to be proved by next quarter's numbers.
Beyond the numbers: the Hugging Face deal and the guarantees
The highlights section mentioned two items with no direct link to the numbers. First, Nvidia announced an agreement to acquire Hugging Face, aiming to scale its platform and make intelligence more open. Hugging Face is the world's largest open-source AI model community; if the deal closes, Nvidia would no longer just sell chips but directly own the place where AI developers gather most. The release disclosed no price or timetable, and details remain to be announced.
Second, the scale of the company's guarantees remains under scrutiny. According to CNBC's account of the call, Nvidia's guarantees tied to customer data center projects now total about $118 billion, including the OpenAI site in Ohio, with exposure the company says would peak at $105 billion. "To date, we have not made any payments related to our guarantees," the company said. A guarantee is not debt, but it ties Nvidia's balance sheet to its customers' construction schedules: if a large customer's project ran into trouble, Nvidia could bear joint liability. It is a topic analysts have pressed on for a year, and one of the numbers most worth tracking in this report.
What this report really answered
Back to the original question: can Nvidia without Huang keep telling the growth story? The report's answer is layered. On demand, the answer is yes — revenue doubled, guidance was raised, and, according to CNBC, the fiscal 2028 growth outlook rose from 70% to 75%; demand for AI infrastructure shows no sign of slowing. On execution, the answer is yes for now — Vera Rubin shipped smoothly, data center revenue grew 110%, and margin slipped only 0.2 points; the change of leadership caused no operational disruption. On leadership, the answer is still blank — whether Kress can lead long-term, whom the board will pick, and whether strategy will change are questions that will take several quarters to resolve.
For the AI industry, the report had a larger significance: Nvidia, as the largest supplier of AI compute, sets the mood for the whole AI sector. This time the report steadied things — a 3% after-hours rise is small, but at least it was not a fall. After November's sell-off, what the market needed most was not a surprise but an absence of bad news. On that measure, Nvidia's report did the minimum required job: proving that the machine still runs normally without Huang, at least this quarter. How long it runs, and how fast, is a question for the next report. What to watch then is clear too: whether gross margin recovers from the 71% level, whether Vera Rubin shipments keep up this quarter's pace, and whom the board picks as permanent CEO. Each of those three answers will reprice the company.
Comments (0)
Sources
Related articles
On 20 November 2026, OpenAI released MedChemBench: 99 medicinal-chemistry problems from 117 papers and patents, covering 761 molecules. GPT-6.2 Astra scores 84.5% overall — but 95.4% on assembling structures and only 50.4% on choosing compounds. The total flatters; the components tell the truth.
On 18 November 2026, Google Research and co-authors posted BuggyMLE to arXiv: bugs injected into 222 working ML repositories, and 14 models slide from 95% on the easiest tier to 9% on the hardest. A 9-billion-parameter model trained easy-to-hard beats Claude Opus 5.5 on the 10-bug tier.
On 20 November 2026, Thinking Machines Lab released Inkling 2 Mini: a 26.4B-parameter looped transformer that runs its middle 12 layers three times per token, fits in about 20GB, and runs offline on a laptop. It leads open-weights models of its size on DeepSWE and SWE-bench Verified — with the caveats printed in the same announcement.