What happened
NVIDIA’s November 17, 2026 earnings release reported third-quarter fiscal 2027 revenue of $115.8 billion for the quarter ended October 25, up 103% year over year and 20% from the previous quarter. Data Center revenue was $107.6 billion, up 110% year over year and 21% sequentially. The company reported GAAP gross margin of 74.8% and diluted earnings per share of $2.70; its non-GAAP figures were 74.9% gross margin and $2.68 per diluted share.
The headline for infrastructure customers is the fourth-quarter outlook. NVIDIA expects revenue of $130.0 billion, plus or minus 3%, for a quarter that includes an additional week. The release states that this outlook “is not assuming any Data Center compute revenue from China.” It does not characterize the assumption as a sales ban, a new export restriction, or a change to availability. It is a stated boundary in the company’s forecast, not a reported operational change for developers.
The release also describes a fast-growing product and deployment roadmap. NVIDIA says it ramped production and began shipping Vera Rubin NVL72, which it calls the fastest product ramp in its history. It names CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius among partners with racks running, and says they are coming online soon at SpaceXSI. It also says Vera Rubin NVL72 systems and Vera CPUs are generally available on Oracle Cloud Infrastructure, and identifies Google Cloud’s Vera Rubin-powered A5X instances as generally available.
Beyond those systems, the company lists infrastructure initiatives across cloud, agent and scientific-computing workloads. It says Anthropic’s access to Grace Blackwell and Vera Rubin capacity was expanded through cloud providers, and that it is collaborating with Anthropic on design and engineering for future architectures. NVIDIA also announced an open agent safety platform and an OpenShell reference architecture, combining OpenShell software, Vera Rubin inference, Vera CPU agent execution and BlueField-4 DPUs running NVIDIA Sentry to monitor agent behavior and enforce policies. These are company-described initiatives; the earnings release does not quantify their adoption or performance.
What the archive adds
New Runtime’s earlier coverage reported AMD’s Q3 2026 data-center revenue at a record $8.5 billion, up 97% year over year (November 4, 2026 note). That was a separate company’s result, not prior coverage of NVIDIA’s China assumption. This is a new fact on the broader AI compute investment story: NVIDIA’s forecast now makes its China revenue assumption explicit.
Trend signal
NVIDIA’s outlook is consistent with the view that scrutiny of its China access will continue even without verified evidence of information transfer (New Runtime analysis). The forecast is not evidence of a newly imposed restriction; it does show the company excluding China Data Center compute from its near-term revenue planning. That distinction matters when interpreting sales expectations against the changing policy environment.
Why now
The assumption appears in NVIDIA’s quarterly earnings outlook alongside a projected $130 billion in next-quarter revenue. Its placement makes the China exclusion part of the company’s stated baseline for the quarter, rather than a separately announced product or policy update.
Why it matters
For AI teams, NVIDIA’s figures reinforce how concentrated current infrastructure spending is in data-center compute: the segment generated $107.6 billion of the quarter’s $115.8 billion in revenue. But this announcement does not tell customers whether a specific GPU will be available to them, at what price, or on what schedule. The China assumption should therefore be read as a forecast boundary, not as a direct operational signal for teams buying cloud capacity elsewhere.
The release also puts agent infrastructure into the same roadmap as accelerators and cloud deployments. NVIDIA describes a reference architecture that separates inference, agent execution and policy monitoring across different components. That is relevant to platform architects as an announced design direction, but the release offers no independent results showing how the system performs in production or what deployment costs it would impose.
Limits and uncertainty
The outlook is a company forecast, not a guarantee of quarterly results. The release gives a total revenue range and says no Data Center compute revenue from China is assumed, but that alone does not establish what China-related sales may actually occur, why the forecast uses that assumption, or how it affects individual products and customers. NVIDIA’s descriptions of Vera Rubin deployments and agent-safety infrastructure are also company statements; the cited financial release does not provide independently verified deployment counts or operational evaluations of the safety architecture.
Reader takeaway
NVIDIA’s near-term revenue outlook is built without China Data Center compute revenue, even as Data Center remains the dominant part of its business. Treat that as a company planning assumption—not proof of a new rule, a sales halt, or changed access to compute. For builders, the practical signal is the scale and pace of infrastructure deployment, while customer-level availability still requires separate confirmation.