The US Commerce Department has approved around ten Chinese firms (including Alibaba, Tencent, ByteDance, JD.com and others) to buy Nvidia's H200, up to 75,000 chips per buyer, along with distributors Lenovo and Foxconn. Three sources told Reuters that not a single delivery has been made. Nvidia CEO Jensen Huang flew with President Trump to Beijing on Wednesday, picked up in Alaska en route to the Xi summit, to try to find a way through.
The interesting fact in that paragraph is the third sentence. For the first time in the four-year history of US chip export controls on China, the binding constraint is not Washington. It is Beijing.
The role reversal
From 2022 through early 2026, every iteration of the export-control story has been the same. The Biden then Trump administrations decided which Nvidia parts could ship to which Chinese customers under which conditions. Chinese firms tried to buy as much as the rules permitted. Nvidia designed products at the regulatory ceiling. Successive cuts (the original A100 and H100 ban, the H800 and A800 down-spec, the dance over L40S and RTX 4090) moved the ceiling. Demand on the Chinese side was assumed throughout.
That assumption no longer holds. Commerce Secretary Howard Lutnick told a Senate hearing last month that "the Chinese central government has not let them, as of yet, buy the chips, because they're trying to keep their investment focused on their own domestic industry." A separate Reuters source said pressure is mounting in Beijing to block or tightly vet the orders. The State Council has just issued two new supply chain security regulations pushing a government-wide effort to identify and eliminate foreign dependencies in critical technology infrastructure. Approved Chinese buyers have read the room and are sitting on their licences.
The 25% tariff workaround is also a security pretext
Part of the deadlock is structural. Trump negotiated an arrangement under which the US takes 25% of the revenue from H200 sales to China. US law does not permit direct export fees, so the workaround routes the chips through US territory before they ship onward to China. This sounds like a clever fiscal trick from Washington's side. From Beijing's side, it looks like the most strategically sensitive chips in the country's data centres will have spent time in US custody on the way in.
The concern about tampering or undisclosed backdoors is not new (Beijing has raised it for years about any imported US technology), but the routed-through-US shipping structure makes it concrete and recent. Whether the security concern is real or pretextual is essentially the wrong question. It is both. The shipping route gives Beijing a defensible reason to do what its industrial policy already wants, which is to redirect demand to domestic suppliers.
The demand-side collapse Huang has already conceded
Huang has said publicly that Nvidia's share of AI accelerators in China has effectively fallen to zero. That figure was hard to credit twelve months ago and is largely uncontested now. Pre-controls, Nvidia held 95% of the Chinese advanced-chip market and China represented 13% of company revenue. Huang estimated the country's AI market at $50 billion this year. The first number is gone. The second is close to gone. The third may turn out to have been the bet Huang was protecting with the H200 carveout, and may already have been lost.
The replacement is not theoretical. DeepSeek, the most internationally visible Chinese AI lab, has spent the last year actively touting its move to Huawei Ascend silicon. Chinese cloud providers have followed. Even where the domestic chips lag Nvidia on raw performance (and they still do at the frontier), the supply-security argument now beats the performance argument inside Chinese procurement. That is the demand shift Huang flew to Beijing to try to reverse. It is not clear what he could offer that would.
The Washington half of the bind
Huang also has a hawkish chorus at home. Chris McGuire of the Council on Foreign Relations, quoted by Reuters, put the case bluntly: "Any deal that allows Nvidia to sell more chips to China means fewer Nvidia chips for US firms, and a smaller US lead in AI over China. It is remarkable that President Trump keeps getting convinced to put Nvidia's interest ahead of America's." That argument carries inside the White House on alternating weeks. The 25% revenue cut is what keeps it from carrying every week.
So the H200 carveout sits in an unstable equilibrium. Washington hardliners want it cancelled. The administration wants the revenue. Beijing does not want the chips delivered. Chinese firms want them but have been told to wait. Nvidia wants the revenue and is being told publicly by its own CEO that the underlying market has already moved past it. The trip with Trump is essentially Huang trying to break the equilibrium in his direction before it breaks in someone else's.
What to watch
Any actual delivery in the next sixty days. A single confirmed shipment to any of the ten approved buyers would signal Beijing has decided to let the H200 in, probably in exchange for something extracted at the Trump-Xi meeting. Continued zero deliveries through summer is the more probable outcome and confirms the market is closed regardless of the licence count.
Huawei Ascend roadmap signals. Huawei's next generation, expected to sample later this year, is the chip that decides whether the domestic-substitution thesis holds at the frontier. If it underwhelms, Chinese hyperscalers may quietly accept the H200 after all. If it ships on or ahead of schedule with credible inference performance, the H200 carveout is moot.
Nvidia's China revenue line in the next two quarterly reports. Whether the H200 carveout produces measurable revenue, or whether it becomes the explicit acknowledgement, in a 10-Q, that the market is gone.
Trump-Xi readouts. Any sign that chip access has been traded for something else (tariffs, agriculture, Taiwan-adjacent posture) reframes the H200 deal as a bargaining chip rather than a commercial product, which is how Beijing has likely viewed it from the start.
