Nvidia Earnings: 7 Essential Takeaways From the $116B Quarter

Nvidia's Q3 fiscal 2027 revenue reached $115.8 billion and guidance rose to $130 billion, in the first report since Jensen Huang resigned. Seven takeaways for anyone building with AI.

This summary is based on Nvidia’s 17 November press release, its earnings call as transcribed by financial outlets and independent analysis. It is not investment advice, and figures are attributed to the sources that reported them.

Nvidia earnings for the third quarter of fiscal 2027 came in at $115.8 billion in revenue, up 103% from a year earlier and 20% from the previous quarter. That beat both Wall Street’s estimate of about $109 billion and Nvidia’s own $108 billion forecast, and the company guided to $130 billion for the current quarter.

The numbers were never really the question this time. Five days earlier, Jensen Huang resigned as chief executive, and Colette Kress ran the call as interim CEO. My read after going through the release and the call: the AI build-out is accelerating, not slowing, and Nvidia’s biggest risks have moved from demand to leadership and geopolitics. Here are the seven takeaways that matter for anyone who builds with AI.

Table of Contents

Nvidia earnings data centre demand
Data centres, not gaming PCs, are now almost the whole story in Nvidia earnings: $107.6 billion of $115.8 billion.

Nvidia Earnings at a Glance

Here are the headline figures for the quarter that ended on 25 October 2026, from Nvidia’s release and the call.

Measure Q3 FY2027 Change
Revenue $115.8 billion +20% quarter on quarter, +103% year on year
Data Centre revenue $107.6 billion +21% quarter on quarter, +110% year on year
Hyperscale About $60 billion +23% quarter on quarter
AI Clouds, Industrial and Enterprise About $47 billion +18% quarter on quarter, nearly tripled year on year
Networking More than $21 billion Not broken out
Edge Computing $8.2 billion +14% quarter on quarter, +42% year on year
Gross margin 74.8% GAAP, 74.9% non-GAAP Above guidance
GAAP net income $65.3 billion +105% year on year
GAAP diluted EPS $2.70 +108% year on year
Q4 revenue guidance $130 billion, plus or minus 3% Includes an extra week

Nvidia now reports revenue by market platform: Hyperscale, AI Clouds, Industrial and Enterprise, and Edge Computing, as Ian Cutress noted in his write-up. That makes these Nvidia earnings harder to compare line by line with older quarters, but the top line tells the story clearly enough.

What Wall Street Expected From Nvidia Earnings

Context matters for any quarter, and this one arrived under unusual pressure. Nvidia’s stock closed at $162.85 on 13 November, about 32% below its 6 October close, according to FourWeekMBA’s tracker. Part of that fall came from a broader sell-off in AI shares. Part came from the leadership news on 12 November.

Going in, Nvidia had guided to revenue of $108 billion, plus or minus 2%, with a 74% gross margin. Analysts had pencilled in a little more, around $109 billion. The previous quarter had already been huge: $96.2 billion in revenue, up 106% year on year, with data centres making up 92.5% of the total.

So the bar was high, and the company cleared it comfortably. Revenue came in nearly $8 billion above its own forecast and about $7 billion above the analyst consensus. Margins beat guidance too. On the numbers alone, these Nvidia earnings were among the strongest the company has ever reported.

A year earlier the comparison looks almost unreal. In the third quarter of fiscal 2026, Nvidia reported $57.0 billion in revenue, $51.2 billion of it from data centres, with a 73.4% gross margin and $31.9 billion of net income. Every one of those lines has roughly doubled since.

1. Nvidia Earnings Show Data Centre Demand Still Doubling

Data Centre revenue reached $107.6 billion, up 110% from a year ago. A year earlier, total company revenue was $57.0 billion. The data centre business alone is now nearly twice the size of the entire company twelve months ago.

The mix is shifting too. Hyperscalers still lead at about $60 billion, but the AI Clouds, Industrial and Enterprise segment nearly tripled year on year to about $47 billion. That is the neoclouds, sovereign AI projects and large companies building their own capacity. Demand is broadening beyond the four or five biggest buyers, which makes it more durable than it looked a year ago.

For anyone worried about an AI spending slowdown, this quarter’s Nvidia earnings are the strongest counter-argument available. Whatever happens to AI stock prices, the companies buying chips are buying more of them.

2. Vera Rubin Had the Fastest Ramp in Nvidia’s History

Vera Rubin, Nvidia’s new platform, generated nearly $20 billion in revenue in its first quarter of production shipments, according to the call. Management called it the fastest product ramp in the company’s history.

Production shipments began in August. Within about three months, Rubin was in production at every major hyperscaler and leading neocloud, and hyperscalers took about one month to reach production, half the time Grace Blackwell needed. On the MLPerf Endpoints benchmark, Nvidia said Vera Rubin delivered ten times more tokens per watt than GB300.

That last figure matters for everyone paying for inference. More tokens per watt means cheaper tokens once the hardware is deployed widely. The roadmap continues with Rubin Ultra next year and Feynman in 2028.

3. Nvidia Earnings Guidance Grew While Margins Shrank

Nvidia guided to $130 billion for the fourth quarter, plus or minus 3%. The quarter includes an extra week and excludes licensed Hopper sales to China. It also raised its growth outlook for fiscal 2028 to about 75%, up from 70%, measured from a higher base.

Margins are heading the other way. Fourth-quarter gross margin is guided to 71% to 72%, which management described as the bottom, down from 74.8% this quarter. Fiscal 2028 margins are guided to 72% to 73%. New platforms usually cost more to build early in their life, and Vera Rubin is ramping fast.

A three-point margin dip on a business this size still leaves extraordinary profitability. But it is the first sign in a while that Nvidia is trading some margin for volume, and it is worth watching in the next set of Nvidia earnings.

4. China Is Now a Risk, Not a Revenue Line

Licensed Hopper data centre sales to China were $2.4 billion in the quarter. Nvidia said it is no longer shipping those products and has excluded them from guidance. New US Commerce Department export controls were released the week of the call, and management said it was still assessing their impact.

The political pressure is real. On 15 November, US Treasury Secretary Scott Bessent asked Nvidia on Fox News to explain whether it knowingly allowed its most advanced chips to reach China, and AFP reported that the Department of Justice is running a criminal investigation. Nvidia also created a National Security and Export Control Oversight Committee.

China used to be a growth market for Nvidia. In this quarter’s Nvidia earnings it reads as a compliance problem. That is a meaningful change in the risk profile, even if the revenue impact is small relative to the whole.

5. Nvidia Earnings Without Jensen Huang

Jensen Huang co-founded Nvidia and led it for more than three decades. On 12 November he resigned as president, CEO and board member after the board disclosed personal conduct it said was inconsistent with company standards. The board accepted the resignation and said it does not intend to reinstate him. Management declined to discuss the circumstances on the call.

Colette Kress, the chief financial officer, is interim CEO, and a new board chairman, Harvey Jones, joined the call. Heidrick and Struggles is leading the search for a permanent chief executive, considering internal and external candidates. Kress told investors it was “business as usual”.

The quarter itself was planned and largely executed before the resignation, so these Nvidia earnings say little about life after Huang. The real test is the next year: product roadmaps, customer relationships and the export-control fight all depended heavily on him. Analysts quoted by Benzinga made the same point, calling leadership a bigger test than any single quarter.

6. The Commitments Behind the Growth

Behind the revenue sits a very large set of obligations. Purchase commitments rose to $390 billion. Data centre lease and cloud service commitments are just over $100 billion, neocloud commitments total $32 billion, and total guarantees reach $118 billion, including $105 billion of credit support for a single campus project.

Nvidia also returned a record $28 billion to shareholders in the quarter, $22 billion through buybacks and $6 billion in dividends, and added $150 billion to its repurchase authorisation, the largest in its history.

And it is investing directly in its customers. Nvidia committed $10 billion to Anthropic’s private placement alongside its IPO. Ian Cutress also noted that income from equity holdings fell from $7.8 billion in the second quarter to $1.8 billion in the third, including a $1.8 billion impairment. The circular nature of some AI investment, where suppliers fund customers who buy their chips, is the critics’ favourite argument, and these numbers give it material.

7. What Nvidia Earnings Mean for AI Builders

If you build with AI rather than trade chip stocks, three things follow from these Nvidia earnings.

Inference should keep getting cheaper. Ten times more tokens per watt on Vera Rubin, deployed across every major cloud within months, puts steady downward pressure on the cost of serving models. Expect more price cuts like Anthropic’s 20% reduction on Fable 5.2.

Capacity is still tight. Demand doubling year on year means rate limits and waiting lists will not disappear soon. Spread your workloads across more than one provider, and design for graceful fallbacks. Our guide to LLM inference servers covers how to run open models yourself as a backstop.

Consumer GPUs are not the priority. The call barely touched gaming, and management said only that it had secured memory. With the memory shortage pushing consumer card prices far above list, local AI builders should not expect relief from Nvidia this quarter. Our best local LLMs guide helps you get the most from the hardware you already have.

If you are Start with Why Link
Paying for API inference Watch for price cuts over the next two quarters Rubin’s efficiency gains feed through to token prices Cut LLM API costs
Running your own models Plan for multi-provider capacity Demand is still outrunning supply Inference servers
Reading the numbers yourself Nvidia’s release and CFO commentary The new segment reporting needs careful reading Nvidia press release

What to Watch Before the Next Nvidia Earnings

The next set of Nvidia earnings, covering the quarter to the end of January, will answer some of the questions this one could not. Here is what I will be watching.

The CEO search. A permanent chief executive, especially an outside hire, would shape strategy for a decade. Any announcement before the next report would matter more than the revenue figure.

The margin floor. Management called 71% to 72% the bottom. If gross margin recovers towards 73% as Rubin matures, the dip was a ramp cost. If it does not, pricing pressure from custom chips may be starting to bite.

Export controls. The new Commerce Department rules and the Justice Department investigation could affect more than China sales if they widen. Nvidia’s new oversight committee will be under scrutiny.

Customer concentration. Hyperscalers are building their own accelerators, from Google’s TPUs to OpenAI’s Jalapeño inference chip. So far that has not dented demand for Nvidia, but the growth of the AI Clouds and Enterprise segment is the best early signal of whether Nvidia can keep diversifying beyond its largest buyers.

Rubin supply. A $20 billion first quarter is a fast start. Whether Rubin can scale further depends on memory, packaging and substrate supply, which is why Nvidia announced new agreements with Kinsus, Samsung Electro-Mechanics and SPIL on this call.

Related on this site: for more context, see Vera Rubin Is Shipping and Nemotron 3 Ultra.

FAQ About Nvidia Earnings

What did Nvidia report for Q3 fiscal 2027?
Revenue of $115.8 billion, up 103% year on year and 20% quarter on quarter, with Data Centre revenue of $107.6 billion. GAAP net income was $65.3 billion and diluted EPS $2.70.

What is Nvidia’s guidance for next quarter?
$130 billion in revenue, plus or minus 3%, including an extra week and excluding licensed Hopper sales to China. Gross margin is guided to 71% to 72%.

Why did Jensen Huang resign?
Nvidia’s board said it disclosed personal conduct inconsistent with company standards. Huang resigned on 12 November and the board does not intend to reinstate him. Management declined to give further detail on the earnings call.

Who is running Nvidia now?
Colette Kress, the CFO, is interim CEO. The board has hired Heidrick and Struggles to find a permanent chief executive.

How much did Vera Rubin contribute?
Nearly $20 billion in its first quarter of production shipments, according to management, the fastest ramp in Nvidia’s history.

Do Nvidia earnings affect GPU prices for consumers?
Not directly this quarter. Consumer card prices are driven largely by the memory shortage, and the call focused on data centre products. Management said only that it had secured memory supply.

Verdict

These Nvidia earnings confirm that AI infrastructure demand is still accelerating. Revenue more than doubled, the new Vera Rubin platform ramped faster than anything before it, and guidance for both next quarter and next year went up. For AI builders, that means cheaper inference over time but continued capacity pressure in the near term.

The caveat is that the risks have changed shape. Margins are dipping as Rubin ramps, China has turned from a market into a legal and political problem, and the company is now looking for its first new chief executive in over thirty years. A strong quarter cannot answer any of those questions.

Your next step: if you pay for AI by the token, check your providers’ pricing pages at the start of each quarter for the next year. The efficiency gains in this report will show up there first, and switching to a cheaper tier or model at the right moment is an easy saving.

Sources and Further Reading

On this site, see how the hardware becomes usable capacity in LLM inference servers compared, plan your own setup with best local LLMs, and keep costs down with LLM API cost optimisation. For the bigger picture, read our AI predictions for 2026.