Nvidia Shipped $2.4B to China, Then Guided Zero. ‘No New Curbs’ Isn’t a Forecast.

Nvidia data center GPU racks beside a map of China, illustrating Nvidia chip export restrictions and zero-China guidance

TL;DR · 30-second read

The Short Version

America’s Treasury secretary says the government does not need to tighten the rules any further on Nvidia selling its chips to China. Nvidia makes the chips that power most artificial intelligence systems.

China still matters to Nvidia. The company shipped $2.4 billion of its older-generation chips there in its latest three months. Yet its forecast for the next three months counts on zero sales of those chips to China.

The takeaway: a promise not to add new rules is not a reliable way to sell. For now, Nvidia is planning as if China buys nothing.

AP reported that the U.S. Treasury secretary said no further restrictions are needed on Nvidia’s chip exports to China. The comment comes days after Nvidia reported record results. In an 8-K filed November 17, the company disclosed $115.8 billion in revenue for the quarter ended October 25, 2026, including $2.4 billion in shipments of Data Center Hopper products to China.

In the same filing, Nvidia guided fourth-quarter fiscal 2027 revenue to $130.0 billion, plus or minus 3%, and stated that it is “not assuming any Data Center compute revenue from China in its outlook.”

Executive Summary

A senior administration official signaling that export curbs on Nvidia should not tighten further is meaningful for a company whose chips sit at the center of the AI buildout. Nvidia’s own numbers, though, put the remark in proportion. The company shipped $2.4 billion of Hopper-generation data center products to China last quarter and then built a $130 billion outlook that assumes none.

That gap is the story. “No further restrictions” describes the direction of future rules. It is not an export license, a volume commitment, or a timeline. Nvidia’s guidance treats China as revenue it cannot count on. Investors, Chinese customers, and the data center operators planning around Nvidia supply are all working with the same uncertainty.

A Ceiling on New Rules Is Not a Floor Under Sales

The useful way to read the Treasury secretary’s comment is against Nvidia’s own planning. According to the CFO commentary filed with Nvidia’s November 17 8-K, shipments of Data Center Hopper products to China totaled $2.4 billion in the third quarter. In the same release, the company guided next-quarter revenue to $130.0 billion, plus or minus 3%, a quarter that includes an extra week. It explicitly assumed zero data center compute revenue from China. Business was flowing to China last quarter, and the company still chose not to count on any of it this quarter.

A pledge of “no further restrictions” does not close that gap. It addresses whether Washington adds new limits. It says nothing about whether existing permissions persist, how quickly licenses are approved, or what volumes are allowed. There is also a structural point. Export licensing for semiconductors has historically been administered by the Commerce Department’s Bureau of Industry and Security, not Treasury. A Treasury secretary’s view signals the administration’s thinking, but it is not the instrument that authorizes shipments. Until something more concrete arrives, such as a formal rule or a durable licensing framework, Nvidia’s guidance is the clearest signal of how the company itself weighs the risk.

For investors, the arithmetic cuts both ways. Any China data center sales in the current quarter would come on top of the outlook rather than being required to meet it. That makes the guidance more conservative on this one line. It also means a supportive policy statement is not, by itself, a reason to revise expectations.

Why $2.4 Billion Is Small, and Why It Still Matters

Against record data center revenue of $107.6 billion, the $2.4 billion in China Hopper shipments is roughly 2% of the segment. China is no longer a pillar of Nvidia’s AI business. Hopper also trails the company’s current lineup. Nvidia says Vera Rubin NVL72 is its fastest product ramp ever and credits Blackwell Ultra for the margin mix. What is moving to China is an earlier generation, which reflects the shape that existing controls have already given this trade.

Small is not the same as irrelevant. A $2.4 billion quarterly run rate is a substantial business for most chip companies. For Chinese AI developers, access to even older Nvidia hardware affects how quickly they can train and serve models. The policy question is less about Nvidia’s headline growth than about whether this residual channel is stable, shrinking, or set to widen. On that question, the company has chosen not to guess in its forecast.

Where the Uncertainty Lands

Nvidia’s growth plan is anchored elsewhere. The earnings release names CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, and Nebius as running Vera Rubin racks. It describes expanded deployments with Meta and expanded capacity access for Anthropic. Supply commitments rose from $279 billion to $390 billion, which Nvidia attributes mainly to multi-year memory procurement. Cloud service and data center lease commitments climbed from $74 billion to $102 billion, with lease terms of up to twenty years. These are long-horizon bets, and the outlook that frames them excludes China. For operators and AI clouds outside China, the near-term demand base Nvidia is guiding to does not depend on how the export debate resolves.

The uncertainty falls hardest on parties that must plan around China specifically. That includes Chinese buyers deciding whether to wait for Nvidia hardware or commit to alternatives, and anyone modeling Nvidia’s upside. The timing adds a governance layer. Nvidia’s November 12 announcement said Jensen Huang resigned as CEO effective November 11, with CFO Colette Kress serving as interim CEO while also remaining CFO. The board said it expects operations, customer commitments, and the product roadmap to continue without interruption. Export policy is one of the files the interim leadership now carries.

The policy debate itself deserves even-handed scrutiny. Supporters of the controls argue that limiting advanced computing protects a national security and technology lead. Critics argue that the controls cost U.S. companies revenue and push Chinese buyers toward domestic suppliers. A single statement does not settle either case. A durable answer would need to show up in rules and licenses, and eventually in Nvidia’s guidance, before buyers can plan around it.

Background

Nvidia designs the graphics processing units (GPUs) and networking systems that power most large-scale AI training and inference, the work of building AI models and then running them for users. Its data center segment generated $107.6 billion of its $115.8 billion third-quarter revenue. The company is currently ramping its Vera Rubin platform across major cloud providers while operating under interim leadership following Jensen Huang’s November 2026 resignation.

Since 2022, the U.S. government has required licenses for exports of the most advanced AI chips to China and has tightened the rules over time. Nvidia has responded by selling earlier-generation or China-specific products where they are permitted. As a result, China has gone from a major market to a small and uncertain share of Nvidia’s data center business, which the company now excludes from its forward guidance.

Sources

Source: Treasury secretary says no further restrictions are needed for Nvidia chip exports to China, AP News report on the administration’s stance on Nvidia exports to China.

Primary sources: NVIDIA 8-K filed November 17, 2026; Exhibit 99.1: NVIDIA Announces Financial Results for Third Quarter Fiscal 2027; Exhibit 99.2: CFO Commentary on Third Quarter Fiscal 2027 Results; NVIDIA 8-K filed November 12, 2026; Exhibit 99.1: NVIDIA Announces CEO Transition.