US Clears Nvidia H200 Sales to Ten Chinese Firms With 25% Revenue Cut, But Beijing Tells Buyers to Stand Down

Overview

The US Commerce Department has cleared roughly ten Chinese companies — including Alibaba, Tencent, ByteDance and JD.com, along with distributors Lenovo and Foxconn — to buy Nvidia’s H200 AI accelerator under a licensing framework first announced last December, The National and Tom’s Hardware reported during President Donald Trump’s Beijing summit with Xi Jinping. Reuters, citing a person familiar with the matter, reported that each approved customer can purchase up to 75,000 chips, a cap independently confirmed by The Next Web.

What hasn’t happened is anything resembling a transaction. As of the summit’s conclusion, “not a single Nvidia H200 chip has shipped to any of the ten approved Chinese buyers,” Tech Times reported. Speaking to reporters aboard Air Force One on the return flight, Trump said Beijing “chose not to” approve the purchases because “they want to develop their own,” per Tom’s Hardware. The framework is in place; the chips are sitting still.

What We Know

The licensing terms

The export framework dates to December 8, 2025, when Trump announced that the United States would permit H200 exports under terms requiring “chips would only be sold to approved U.S. customers and that 25 percent of the revenue would go to the U.S. government as a tax,” according to the Council on Foreign Relations. To execute that arrangement, every chip must physically pass through US territory for third-party inspection before re-export — a requirement Tom’s Hardware describes as routing each H200 through US territory “for third-party inspection before re-export,” with Nvidia required to “remit 25% of sale revenues to the U.S. Treasury.”

The ten cleared buyers fall into two groups, per Reuters: hyperscalers and platform operators (Alibaba, Tencent, ByteDance, JD.com) and distributors (Lenovo, Foxconn) through whom additional Chinese customers may route purchases. The Next Web summarized the state of play in one line: “Order books exist, but the deliveries do not.”

Beijing’s counter-pressure

The stall is not a market failure — it is policy on the Chinese side. Commerce Secretary Howard Lutnick told reporters that China has actively “blocked imports in an effort to steer investment toward domestic chipmakers like Huawei,” per Tom’s Hardware. The Next Web reported that “Chinese buyers have held back after Beijing told domestic technology firms to pause H200 orders earlier this year,” with regulators running a “parallel supply-chain security review aimed at reducing dependence on US chips.”

The directive has bite because Chinese hyperscalers now have a domestic alternative. On Alibaba’s May 13 earnings call, an executive declared that the company’s T-Head chip unit had achieved “scaled mass production” of proprietary GPUs, according to The News International. The same executive framed the trade-off in margin terms: “In an environment of compute scarcity, this structural advantage is favourable to our revenue growth and gross margin improvement.” Alibaba CEO Eddie Wu told analysts on the same call, “I can tell you that today there isn’t a single card on our service that is idle,” the South China Morning Post reported.

Tencent’s first-quarter capital expenditure reached 31.9 billion yuan — a 63 percent increase from the prior quarter, per the SCMP — and Chief Strategy Officer James Mitchell told analysts the company will “dramatically increase capital spending, particularly in the second half of the year” as China-designed GPUs become “available to us month by month,” according to The News International. Goldman Sachs now forecasts Tencent capex at 165 billion yuan in 2027, more than double 2025 levels, the SCMP reported.

The Air Force One delegation

Jensen Huang’s late addition to Trump’s Beijing delegation underscored how much Nvidia has riding on a thaw. Trump “picked him up in Alaska en route to a summit,” per Reuters. According to PBS NewsHour, the executive contingent traveling on Air Force One included Tesla and SpaceX CEO Elon Musk, Apple’s Tim Cook, BlackRock’s Larry Fink, Blackstone’s Stephen Schwarzman, Cargill’s Brian Sikes, Citi’s Jane Fraser, Goldman Sachs’ David Solomon, GE Aerospace’s H. Lawrence Culp, Micron’s Sanjay Mehrotra, Qualcomm’s Cristiano Amon, Visa’s Ryan McInerney, Illumina’s Jacob Thaysen, Mastercard’s Michael Miebach, Coherent’s Jim Anderson, Meta’s Dina Powell McCormick, and Boeing’s Kelly Ortberg. The Next Web wrote that Huang’s mission was to seek “a green light from Beijing for the H200 deliveries already cleared in Washington, and a reciprocal easing on Chinese export curbs.”

He didn’t get it. Trump told reporters at the summit’s close that “a lot of different problems were settled,” per Tech Times, but no signed bilateral AI governance document emerged and rare-earth exports remain roughly 50 percent below pre-restriction baseline levels. Treasury Secretary Scott Bessent told reporters that officials were exploring “some sort of emergency communication system” to prevent weaponization of powerful AI systems, also per Tech Times.

Analysis

For Nvidia, the licensing framework is simultaneously a partial victory and a structural problem. Huang has repeatedly argued that the company’s case for engagement with China is about more than near-term sales. The National quoted him as contending that “the benefits of spreading US technology outweigh the dangers some have warned about.” That argument is harder to make when China’s official policy is now visibly aligned against the deal his company just won the right to make. Tom’s Hardware reported that Nvidia’s market share in China has fallen “from approximately 95% to what CEO Jensen Huang described as ‘essentially zero,’” with analysts estimating that a functional export framework could restore $3.5–4 billion in annual revenue. The Next Web put the broader exposure at “roughly 5% of Nvidia’s revenue from a pre-controls high above 20%.”

For critics of the policy, the licensing volume is the alarm. Chris McGuire of the Council on Foreign Relations told The National that the framework, if executed, “would triple the amount of AI computing power that China adds next year.” A ten-buyer, 75,000-chip-each cap implies an upper bound of 750,000 H200s — a number that exists on paper while Beijing’s domestic-procurement push works to ensure it never materializes. Past Machine Herald coverage has tracked the parallel push: Zhipu AI’s GLM-5 became the first frontier model trained entirely on Chinese chips, and Alibaba’s RISC-V division unveiled the Xuantie C950 processor in March, both pointing to a domestic stack that no longer treats Nvidia silicon as essential.

The political economy of the 25 percent revenue arrangement is itself unusual. Because US law does not permit direct export taxes, the chips must be routed through US territory and treated as a re-export, per Reuters. The structure means that every H200 sold under the framework generates a Treasury payment tied to a single private company’s pricing — a precedent that has no recent analog in US technology trade. With Beijing currently blocking the orders, the Treasury collects nothing; with Beijing relenting, Nvidia would be transferring a quarter of each unit’s revenue to Washington while shipping accelerators that competitive pressure has already pushed analysts to view as the second-tier option behind Blackwell.

What We Don’t Know

• Whether Beijing will relent. The pause is informal — a directive to Chinese buyers, not an explicit ban — and could in principle be lifted. But the visible parallel investment in Huawei Ascend and Alibaba T-Head suggests the policy goal is substitution rather than dual-tracking.
• Whether the 75,000-per-buyer cap is firm. Only Reuters and The Next Web have published the figure; the licenses themselves are not public. The cap may be a ceiling rather than an allocation.
• Whether other US chipmakers fall under the same framework. The original December announcement implied that “other tech companies like Intel would be offered a similar opportunity,” per the CFR, but no other firm has had its terms publicly described.
• What happens at the next renewal point. The current US-China trade truce — which reduced US tariffs from 145 percent to 30 percent and Chinese tariffs from 125 percent to 10 percent, per Tech Times — expires in autumn 2026. A breakdown there would likely take the H200 framework with it.

For now, the headline numbers describe a market that exists only in the licensing files. Ten buyers, six distributors, a 25 percent Treasury share, 75,000-chip caps, and zero deliveries.