Nvidia says Vera Rubin NVL72 has begun shipping, with racks already running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. The hardware update accompanied fiscal third-quarter data-center revenue of $107.6 billion, up 110% from a year earlier.
The company’s November 17, 2026, earnings release reported total revenue of $115.8 billion for the quarter ended October 25. Nvidia also forecast $130 billion in fourth-quarter revenue, plus or minus 3%, explicitly excluding any assumed China data-center compute revenue.
These are Nvidia’s first quarterly results since Jensen Huang’s departure, with Colette Kress serving as interim CEO while retaining her finance responsibilities. The operating story deserves attention on its own: Nvidia is reporting another substantial increase in infrastructure sales while bringing a new computing platform into cloud deployments.
Three details make this report particularly useful for assessing the AI industry: how much of Nvidia’s business now depends on data centers, what its deployment language says about access to Vera Rubin, and the conditions behind that next-quarter forecast.
Data Centers Now Generate Nearly 93% of Revenue
Nvidia’s data-center business grew faster than the company overall, both sequentially and year over year. Its official results show the scale:
| Metric | Q3 FY2027 | Change From Previous Quarter | Change From Year Earlier |
|---|---|---|---|
| Total revenue | $115.8 billion | +20% | +103% |
| Data-center revenue | $107.6 billion | +21% | +110% |
| GAAP operating income | $77.5 billion | +22% | +115% |
| GAAP net income | $65.3 billion | +9% | +105% |
| GAAP gross margin | 74.8% | Down 0.2 percentage points | Up 1.4 percentage points |
Dividing reported data-center revenue by total revenue puts the segment at approximately 92.9% of Nvidia’s sales. Everything outside that segment contributed a combined $8.2 billion.
That concentration changes how to read Nvidia’s results. A product announcement elsewhere in the business can still matter, but data-center spending now determines most of the company’s financial direction. Cloud operators’ ability to finance, power and deploy infrastructure is consequently central to Nvidia’s growth.
The sequential figures are especially useful. Total revenue rose from $96.2 billion to $115.8 billion, adding $19.6 billion in one quarter. Data-center revenue grew 21% over the same period, showing that the business expanded substantially even against its already large preceding quarter.
I’d give those sequential gains more weight than the doubling headline alone. Year-over-year growth captures the scale of the expansion; another roughly one-fifth increase in three months shows how quickly Nvidia’s revenue base is still changing.
The release does not break out Vera Rubin revenue, however. The overall data-center increase cannot be assigned to the new platform.
Vera Rubin Has Shipped, but Cloud Availability Varies
Nvidia describes Vera Rubin NVL72 as the fastest product ramp in its history. More concretely, it says production has ramped, shipments have begun and racks are running at named partners.
The release distinguishes several deployment milestones:
- CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius: Nvidia says Vera Rubin racks are running.
- Oracle Cloud Infrastructure: Nvidia announced general availability of Vera Rubin NVL72 systems and Vera CPUs.
- Google Cloud: Nvidia announced general availability of Vera Rubin-powered A5X instances.
- SpaceXSI: Nvidia says racks are coming online soon, placing it at a different stage from the running deployments.
Those distinctions matter for developers and infrastructure buyers. A rack operating inside a provider’s infrastructure is a deployment milestone. General availability is a customer-access milestone. Neither statement alone specifies available regions, account eligibility, capacity quotas or how quickly a customer can secure a large allocation.
My reading is that the general-availability announcements are the most actionable details for prospective cloud customers. They identify services to investigate, while the broader partner list establishes where Nvidia says deployment is underway.
Shipment status, ramp speed and partner operation remain Nvidia-reported claims. The available evidence does not provide independent utilization measurements, shipment counts, workload results or deployment dates detailed enough to compare partners’ rollout speeds.
Kress’s statement that “our roadmap is on track” is therefore best read alongside the named deployments, not as a substitute for them. Those deployments provide a more concrete sign of execution, while leaving important questions about scale and customer access unanswered.
The release also offers no basis for calculating Vera Rubin’s cost per training run or inference request. Buyers assessing competition between AI platforms still need workload-specific performance, pricing and availability information.
The $130 Billion Outlook Has Two Important Conditions
Nvidia expects fourth-quarter revenue of $130 billion, plus or minus 3%. That produces a forecast range of approximately $126.1 billion to $133.9 billion.
At the midpoint, revenue would increase by $14.2 billion, or about 12.3%, from the third quarter. But the outlook includes an additional week, so that percentage is not a comparison between equal-length reporting periods.
The extra week is easy to overlook. It gives Nvidia more time to recognize revenue, meaning the headline sequential increase should not be treated as an equivalent increase in weekly sales activity. The release does not provide enough detail to calculate an exact underlying weekly growth rate.
China Compute Revenue Is Excluded, Not Every China Sale
Nvidia explicitly says it is assuming no data-center compute revenue from China in its forecast. That is narrower than saying it expects no revenue of any kind from China, and it is a forecasting assumption rather than a final sales result.
CTech’s earnings coverage places that exclusion in the context of restrictions affecting Nvidia’s access to the Chinese market.
The practical implication is significant: Nvidia’s $130 billion midpoint does not depend on China data-center compute sales materializing. It is evidence of management’s confidence in revenue from other markets and categories, although the release does not provide the geographic breakdown needed to map that demand precisely.
I would treat the exclusion as a clearer statement of the forecast’s boundary, not as proof that China no longer matters. The outlook says what Nvidia has assumed for one quarter. It does not settle the longer-term commercial consequences of restricted access.
Higher Revenue Comes With Lower Expected Margins
Nvidia forecasts both GAAP and non-GAAP gross margins between 71% and 72%, compared with third-quarter figures of 74.8% and 74.9%, respectively.
For GAAP gross margin, that implies a sequential decline of 2.8 to 3.8 percentage points. Revenue growth and margin direction are therefore sending different signals: Nvidia expects to sell more, while retaining a smaller share of each sales dollar after cost of revenue.

The release does not isolate the causes of that expected decline. It would be premature to assign it entirely to Vera Rubin’s ramp, component costs or product mix. Still, the forecast is a useful reminder that record sales do not automatically mean improving unit economics.
Vera CPUs Broaden the Platform Story
Frequently Asked Questions
3 questions
1Is Nvidia Vera Rubin Shipping Now?
Nvidia said on November 17, 2026, that Vera Rubin NVL72 shipments had begun and racks were running at five named cloud partners. Availability still depends on the provider and service.