Skip to content
MARKET NOTES

NVIDIA's Q3 FY27 Beat: The Customer Mix the Models Missed

NVIDIA reported $115.8B of revenue, $6.7B above consensus. The surprise was in who bought: AI clouds, industry and enterprise beat one widely used model by $17.8B while hyperscalers fell $11.9B short. Adjusted for an extra week, the Q4 guide implies slower growth than the headline.

Parth PatelPUBLISHED UPDATED 6 MIN READ

NVIDIA reported third-quarter revenue of $115.8 billion on 17 November, up 103% from a year ago. Within hours, the most widely shared reading of the quarter was a line from AlphaSense: revenue beat by $6.5 billion, and the customer mix is what the model missed. AI Clouds, Industrial & Enterprise beat the Canalyst model by $17.8 billion, while Hyperscale missed by $11.9 billion.

Both numbers are right. I want to look at what they mean, because a beat is always a beat against something, and the something matters more than the size. I've gone back to NVIDIA's own filings for every figure below, and used the third-party estimates only where they're the point.

Q3 FY27 against expectations

Beat

$115.8B

Revenue

+$6.7B vs LSEG consensus

Beat

$107.6B

Data Center revenue

+$5.8B vs consensus

Above guide

74.9%

Non-GAAP gross margin

guided 74.0% ± 0.5 pt

Beat

$2.68

Non-GAAP EPS

vs $2.48 LSEG consensus

Source: NVIDIA Q3 FY27 press release (17 Nov 2026); consensus as compiled on AlphaSense's public dashboard (LSEG via CNBC; Visible Alpha via S&P Global).

A beat against what?

There are at least three benchmarks for the quarter, and they give three different answers:

  1. NVIDIA's own guidance, issued in August: $108.0 billion, plus or minus 2%. Revenue beat the midpoint by $7.8 billion and the top of the range by $5.7 billion.
  2. Consensus, the average analyst estimate just before the print: about $109.1 billion according to LSEG. The beat was $6.7 billion.
  3. The Canalyst model, a detailed forecast model last updated on 26 August: $109.3 billion. The beat was $6.5 billion, the figure in the headline.

Q3 cleared the top of its guidance range; Q4 is guided above consensus

Revenue, US$ billions: company guidance range and midpoint, consensus, the Canalyst model and the reported figure

  • Q3 FY27 revenue (reported)

    Reported $115.8B
  • Q4 FY27 revenue (guided)

  • Company guidance range
  • Midpoint
  • Consensus
  • Model forecast
  • Reported
Q3 cleared the top of its guidance range; Q4 is guided above consensus
Q3 FY27 revenue (reported)Company guidance range $105.8B to $110.2B; Guidance midpoint $108.0B; LSEG consensus $109.1B; Canalyst model, 26 Aug $109.3B; Reported $115.8B
Q4 FY27 revenue (guided)Company guidance range $126.1B to $133.9B; Guidance midpoint $130.0B; LSEG consensus $125.1B; Canalyst model, 26 Aug $118.9B

Q4 includes an additional week. Canalyst figures are from the model version dated 26 Aug 2026, before the Q3 print.

Source: NVIDIA Q2 and Q3 FY27 press releases; consensus and Canalyst figures as shown on AlphaSense's public dashboard (18 Nov 2026).

The differences are small next to the beat itself, so I don't want to make too much of them. But they set up the more interesting question. Consensus and the model were within $0.2 billion of each other on total revenue. Where they were wrong, they were wrong in the same direction. The question is where inside the total the surprise came from.

Where the surprise was

NVIDIA splits Data Center revenue by customer type. Hyperscale is the largest cloud and internet companies. AI Clouds, Industrial & Enterprise, which NVIDIA abbreviates ACIE, covers specialist AI clouds ("neoclouds"), AI-native companies, enterprises and governments building sovereign AI capacity. Here is how each came in against the Canalyst model.

The beat came from AI clouds and enterprise, not the hyperscalers

Q3 FY27 reported revenue minus the Canalyst model forecast, US$ billions

  • AI Clouds, Industrial & Enterprise+$17.8B ($47.5B vs $29.7B)
  • Edge Computing+$0.7B ($8.2B vs $7.5B)
  • Hyperscale−$11.9B ($60.1B vs $72.0B)
  • Total revenue+$6.5B ($115.8B vs $109.3B)
The beat came from AI clouds and enterprise, not the hyperscalers
AI Clouds, Industrial & Enterprise+$17.8B ($47.5B vs $29.7B)
Edge Computing+$0.7B ($8.2B vs $7.5B)
Hyperscale−$11.9B ($60.1B vs $72.0B)
Total revenue+$6.5B ($115.8B vs $109.3B)

Segment differences use NVIDIA's rounded figures, so they sum to $6.6B rather than $6.5B. The model is dated 26 Aug 2026.

Source: NVIDIA Q3 FY27 CFO commentary; Canalyst forecasts as shown on AlphaSense's public dashboard. Author's calculations.

On its face, that is a large miss on the biggest customers, more than offset by a much larger beat elsewhere. But one detail made me look harder. The model, dated after NVIDIA's second-quarter results, forecast ACIE at $29.7 billion for the third quarter. NVIDIA had just reported ACIE at $40.3 billion for the second. The model was effectively expecting that segment to shrink by about a quarter in three months, while NVIDIA's own figures showed it growing 25% sequentially.

I can't see inside the model, so I can't say why. One possibility is NVIDIA's reclassification in the second quarter, when it moved one customer from ACIE to Hyperscale and restated earlier quarters; a model built on the old split would put too much in Hyperscale and too little in ACIE. Whatever the cause, the lesson is general. A headline that says the model missed partly tells you how old and how well-mapped the model was.

The mix over time

So the more useful question is what NVIDIA's own figures show. Here are the five quarters it has published on the current, restated basis.

Both customer groups have grown fast; ACIE is now 44% of Data Center

NVIDIA Data Center revenue by customer type, US$ billions

$41.1B
$51.2B
$75.2B
$89.0B
$107.6B
  • Hyperscale
  • AI Clouds, Industrial & Enterprise
Both customer groups have grown fast; ACIE is now 44% of Data Center
PeriodHyperscaleAI Clouds, Industrial & EnterpriseTotal
Q2 FY26$24.2B$16.9B$41.1B
Q3 FY26$35.1B$16.1B$51.2B
Q1 FY27$43.1B$32.2B$75.2B
Q2 FY27$48.7B$40.3B$89.0B
Q3 FY27$60.1B$47.5B$107.6B

Q4 FY26 is omitted because NVIDIA has not published it on the restated basis that followed its Q2 FY27 reclassification.

Source: NVIDIA CFO commentaries for Q2 FY27 (26 Aug 2026) and Q3 FY27 (17 Nov 2026).

Table 1: ACIE's share of Data Center revenue

QuarterHyperscaleAI Clouds, Industrial & EnterpriseACIE share of Data CenterHyperscale growth, Q/QACIE growth, Q/Q
Q2 FY26$24.2B$16.9B41%n/an/a
Q3 FY26$35.1B$16.1B31%+45%−5%
Q1 FY27$43.1B$32.2B43%n/an/a
Q2 FY27$48.7B$40.3B45%+13%+25%
Q3 FY27$60.1B$47.5B44%+23%+18%

Source: NVIDIA CFO commentaries for Q2 and Q3 FY27, restated basis. Shares and growth rates are the author's calculations; quarter-on-quarter growth is not shown where the previous quarter isn't published on the same basis.

Two things stand out. First, the dramatic year-on-year comparison, ACIE nearly tripling, is partly a story about a weak base: a year ago ACIE was 31% of Data Center revenue, an unusually low share. In all three quarters published since on the same basis, it has been between 43% and 45%. Second, in the most recent quarter Hyperscale actually grew faster, 23% against 18%. The mix didn't lurch this quarter. What changed is that a forecast had assumed it would.

The part that deserves attention

If the mix is broadly steady, the question worth asking is about the quality of the ACIE side. It is the part of NVIDIA's business most exposed to customers that depend on outside financing. NVIDIA is unusually direct about this in its CFO commentary: neoclouds and frontier labs "are growing faster than their balance sheets and long-term credit profiles can support." In response, it has entered into arrangements to help select customers secure land, power and data center capacity. It discloses up to $32 billion of future neocloud commitments, which take effect only if those customers secure financing to buy NVIDIA's infrastructure. Separately, its supply commitments rose from $279 billion to $390 billion in the quarter, mostly for memory.

None of this is hidden, and none of it is necessarily a problem. But it is the pattern Howard Marks described in his recent memo on AI, which we looked at earlier this month: circular deals among chipmakers, clouds and labs, and a predictability score of 2 to 3 out of 10 for speculative neocloud data centers. When the fastest-growing customer group is also the one its supplier describes as stretched, growth and risk are arriving through the same door.

Reading the Q4 guide

NVIDIA guided fourth-quarter revenue to $130.0 billion, plus or minus 3%, about $5 billion above the roughly $125 billion consensus. The headline growth is 12.2% over the third quarter. But the release notes that the fourth quarter includes an additional week: 14 weeks against 13.

Adjusting the Q4 guide for its extra week

+12.2%

Q4 guide vs Q3, as reported

$130.0B vs $115.8B

+4.2%

Per week, Q4 guide vs Q3

$9.29B vs $8.91B a week

Source: Author's calculation from NVIDIA's Q3 revenue and Q4 outlook: Q3 was 13 weeks (27 Jul to 25 Oct 2026), Q4 is 14.

Per week, the guidance midpoint implies revenue of about $9.29 billion against $8.91 billion in the third quarter, an increase of about 4%. The bottom of the range implies about 1%, the top about 7%. Weekly revenue is a rough measure, since shipments don't arrive evenly, but it changes the impression. On a like-for-like basis, the guide describes a business still growing, more slowly than the headline suggests, with gross margins guided down to 71% to 72% from 74.9% as the Vera Rubin ramp begins.

Where this could be wrong

  • The model comparison is one model. Canalyst is a single forecast, and its 26 August date means it hadn't absorbed any information from the quarter. Other models may have split the customers very differently.
  • The extra-week arithmetic is crude. Revenue isn't spread evenly across weeks, and NVIDIA doesn't say how much the 14th week contributes. The adjustment shows the direction of the effect, not its precise size.
  • Customer categories are NVIDIA's. It has already reclassified one customer this year, and the line between a hyperscaler and an AI cloud is a matter of definition. Shares could shift with the next restatement.
  • Consensus figures are second-hand. I took them from coverage compiled on AlphaSense's public dashboard, not from the data providers directly.
  • Commitments are contingent. The $32 billion of neocloud commitments may never be drawn in full. That is the point of the disclosure, but it also means the eventual figure is unknown.

Closing thoughts

Earnings season rewards a fast verdict, and "beat by $6.5 billion" is a fast verdict. The slower reading is more useful. The size of the beat depends on what it's measured against. The customer mix the model missed was largely visible in NVIDIA's previous quarter. The next quarter's growth is flattered by its calendar. And the part of the business growing fastest is the part most dependent on other people's financing. None of that is a judgment on the company. It is a reminder that the number in the headline is the beginning of the analysis, not the end of it.

This article is educational and is not personalized financial advice, and it is not a recommendation to buy, sell or hold any security. Investing involves risk of loss, and past performance does not predict future results. Company figures are from NVIDIA's filings; third-party estimates are attributed; the analysis is the author's own. Data as of 21 November 2026.

How much did NVIDIA beat revenue expectations in Q3 FY27?

NVIDIA reported $115.8 billion of revenue for the quarter ended 25 October 2026. That was $7.8 billion above its own guidance midpoint of $108.0 billion, about $6.7 billion above LSEG consensus of $109.1 billion and $6.5 billion above the Canalyst model's $109.3 billion forecast.

What is NVIDIA’s AI Clouds, Industrial & Enterprise segment?

It is one of NVIDIA's two Data Center customer groups, alongside Hyperscale. It covers specialist AI cloud providers (neoclouds), AI-native companies, enterprises and sovereign customers. In Q3 FY27 it reported $47.5 billion, up 195% from a year ago and 44% of Data Center revenue.

What did NVIDIA guide for Q4 FY27?

Revenue of $130.0 billion, plus or minus 3%, with gross margins of 71% to 72% and no Data Center compute revenue from China assumed. The quarter includes an extra week; adjusted per week, the midpoint implies about 4% growth over Q3 rather than the reported 12.2%.

Sources & references

  1. NVIDIA Announces Financial Results for Third Quarter Fiscal 2027 (17 Nov 2026), NVIDIA
  2. CFO Commentary on Third Quarter Fiscal 2027 Results (Form 8-K, Exhibit 99.2), NVIDIA via SEC EDGAR
  3. CFO Commentary on Second Quarter Fiscal 2027 Results (Form 8-K, Exhibit 99.2), NVIDIA via SEC EDGAR
  4. Nvidia (NVDA): Q3 FY27 earnings dashboard (consensus and Canalyst model figures), AlphaSense SuperAnalyst

The materials on this website are for illustration and discussion purposes only and do not constitute an offering. An offering may be made only by delivery of a confidential offering memorandum to appropriate investors. PAST PERFORMANCE IS NO GUARANTEE OF FUTURE RESULTS. Disclosures

Investing in the US, Canada, India, China and Japan involves currency, emerging-market, liquidity and regulatory risks. International and emerging-market risk

SHARELINKEDINXEMAIL

More in Market Notes

View all →

Market Notes · Nov 6, 2026 · 7 MIN

Rates Are Rising Again: Where Credit Still Offers Value

The Fed hiked to 3.75%–4.00% and the 10-year briefly topped 5%, for different reasons at each end of the curve. Short-duration credit lets you collect the higher yield without forecasting the hard part. Here is how higher rates reshaped high yield, loans, RMBS and CLOs, and where value remains.

Sumit Pande

Market Notes · Nov 5, 2026 · 6 MIN

Nifty 50 Flat for 2.5 Years: Why Long-Term Investors Quit

The Nifty 50 returned 0.20% in the 29 months from May 2024 to October 2026, after rising about 17% and falling about 16% along the way. Here is why roughly 2.5 years without a gain is where long-term investors give up, what quitting cost, and how to plan for it.

Parth Patel

Factor Research · Nov 17, 2026 · 6 MIN

Dan Benton's 1991 Tech Rules: What Held Up and What Didn't

Dan Benton's 20 rules for technology investing, written at Goldman Sachs around 1991, still circulate. In 2026 he rewrote them. The rules about earnings expectations held up; the summer-seasonality rule fit the six summers he had seen and then faded, as 40 years of Nasdaq-100 data show.

Parth Patel

Want to discuss this research?

Talk to us