NVIDIA announced its fiscal 2027 third-quarter results on November 17, 2026, U.S. time. Revenue rose 103% year over year to $115.8 billion, $7.8 billion above the midpoint of the company's own outlook given last quarter. Data center revenue was $107.6 billion, about 92.9% of total revenue. In terms of year-over-year dollar growth, the segment covering AI cloud providers and industrial and enterprise customers outpaced hyperscale customers. The quarter ended on October 25, before founder Jensen Huang resigned. What the interim leadership inherits is a business that, alongside selling hardware, handles long-term component procurement and credit guarantees tied to customers' capital spending.
The customer group behind most data center growth
Data center revenue rose 110% year over year. According to the CFO commentary attached to NVIDIA's SEC filing, revenue from hyperscale customers, including the large cloud providers, was $60.1 billion, still the larger segment by dollar amount. AI cloud, industrial and enterprise customers contributed $47.5 billion, but their year-over-year growth rate of 195% exceeded the 71% growth in hyperscale.
| Data center customer group | Fiscal 2026 Q3 | Fiscal 2027 Q3 | Year-over-year increase |
|---|---|---|---|
| Hyperscale | $35.1 billion | $60.1 billion | $25.0 billion |
| AI cloud, industrial and enterprise | $16.1 billion | $47.5 billion | $31.4 billion |
| Total | $51.2 billion | $107.6 billion | $56.4 billion |
Note: Based on NVIDIA's earnings presentation and the CFO commentary. Increases are differences between the rounded figures NVIDIA published.
AI cloud, industrial and enterprise customers accounted for about 56% of the year-over-year increase in data center revenue: $31.4 billion divided by $56.4 billion, the sum of the two segments' increases. The segment made up about 44% of data center revenue this quarter. NVIDIA discloses figures only by customer group; it does not publish revenue by individual customer or a breakdown of end users who access its products through the cloud.
By product, revenue from the new compute platform Vera Rubin rose by nearly $20 billion from the previous quarter. Total company revenue grew $19.6 billion over the same period (from $96.2 billion to $115.8 billion), roughly the same scale as the Vera Rubin increase. According to NVIDIA, racks built around Vera Rubin are running at CoreWeave as well as at Google Cloud and Microsoft Azure.
On the November 17 earnings call, interim CEO and CFO Colette Kress cited two reasons AI cloud, industrial and enterprise revenue grew from the prior quarter. One was NVIDIA's shipment of Vera Rubin to major AI-specialized cloud providers. The other was that developers of frontier models have begun building their own data centers. The customer-group table is year over year, while the Vera Rubin figure reflects the quarter-over-quarter change by product.
Multiyear memory purchases and credit guarantees for capital spending
The value of NVIDIA's procurement commitments to secure component supply and production capacity rose from $279.0 billion at the end of the previous quarter to $390 billion. The $111 billion increase came mainly from multiyear memory purchases. Cloud service agreements and leases for data centers not yet in use together rose from $74 billion to $102 billion. NVIDIA also uses that computing capacity to design and test chips and networking, and to develop its own AI models such as Nemotron.
In its agreements with AI-specialized cloud providers, NVIDIA took on conditional obligations of up to $32 billion. The conditions are that the provider secures funding to buy the equipment and brings the facilities into operation, and that NVIDIA accepts the delivery. NVIDIA earns revenue from the initial equipment sale. If certain criteria are also met, it receives a share of the revenue the provider earns from serving third parties. Delays in financing or acceptance could change the amount or timing of NVIDIA's obligations, and in some cases part of the obligation could fall away. Whether the equipment NVIDIA delivers comes online and whether the provider can win customers will determine the revenue-sharing income.
In the event of a default or similar, the total NVIDIA could be required to pay under its credit guarantees is up to $118 billion. Of that, up to $105 billion supports guarantees for SB Energy's data center project in Ohio, covering the securing of land and power and the construction of building shells. OpenAI will use the site under a 20-year lease. The plan calls for about 4.25 GW across nine buildings, with the guarantee starting as each building is completed and ready for use. The first guarantee is expected to begin in fiscal 2029, and no payments had been made under the guarantees at the time of disclosure.
A next-quarter outlook that excludes China compute products
In the third quarter, NVIDIA shipped $2.4 billion worth of Hopper-generation data center products to China. On the earnings call, Kress said the company has already stopped shipping licensed Hopper products to China. The fourth-quarter outlook includes no revenue from data center compute products in China. The third-quarter outlook given at the last earnings report rested on the same assumption.
The fourth-quarter revenue outlook is $130 billion, plus or minus 3% ($126.1 billion to $133.9 billion), and the midpoint is $14.2 billion above the third-quarter result. For accounting purposes, the fourth quarter has one more week than the third. NVIDIA expects gross margin of 71% to 72% on both a U.S. GAAP and a non-GAAP basis. Against this quarter's GAAP result of 74.8%, even the top of that range is 2.8 percentage points lower.
GAAP operating income this quarter rose 115% year over year to $77.5 billion, and net income rose 105% to $65.3 billion. Gross margin rose from 73.4% a year earlier and dipped slightly from 75.0% in the previous quarter. NVIDIA attributes the year-over-year improvement to the Blackwell Ultra product mix and the sequential decline to the ramp of Vera Rubin. Kress said strong demand and additional secured supply led the company to raise its fiscal 2028 revenue growth outlook from 70% to about 75%. It expects gross margin of 72% to 73% for that year.
Export controls and open models after Huang's resignation
NVIDIA's board accepted Huang's resignation as president, CEO and director on November 11, and the company announced it the next day in a personnel disclosure. The board named Kress interim CEO and is selecting a permanent successor. On the earnings call, Chair Jones said the board accepted the resignation unanimously and that it has no intention of reappointing Huang as CEO.
On November 17, the same day as the results, the board announced it had established a standing committee to oversee national security and export controls. The internal team responsible for export controls will report directly to both the committee and the interim CEO. Through the committee, which will meet at least once a quarter, the board has created a channel to receive export control reports directly.
On the call, Kress cited examples of companies further training models on their own data and adapting them to their operations, and indicated that NVIDIA will continue investing in open models. In September, NVIDIA agreed to acquire Hugging Face, and the acquisition process was still under way when the results were announced. When announcing the agreement, NVIDIA said Hugging Face, which is used to distribute models, would continue to support multiple clouds and AI accelerators from multiple companies.
If companies further train open models under their license terms and build them into everyday operations, they will need computing resources not only for training but also for inference, which keeps the models running. Because there are multiple supported platforms, which platform captures that demand as revenue will depend on where companies choose to run both training and inference.
