This is the president's deal, and not everyone agrees with it.
Somewhere between national security and economic pragmatism, the Trump administration is attempting to thread a needle that may not exist. A proposed cap of 75,000 Nvidia H200 chips per Chinese firm represents neither the openness the market desires nor the restriction security hawks demand — a compromise born of months of negotiation, rejection, and recalibration. At its core, this is a question as old as great-power rivalry: whether controlled engagement serves a nation's interests better than embargo, and who ultimately gets to decide.
- Chinese tech giants like Alibaba and ByteDance want more than twice the 75,000-chip ceiling being proposed, creating immediate tension between what buyers need and what Washington will allow.
- The H200 itself is already a concession — six times more powerful than what was previously permitted into China, arrived at only after Beijing rejected a lesser chip and the White House retreated from offering its most advanced hardware.
- A ban on using purchased chips in overseas data centers has quietly closed one of the most commercially valuable use cases, leaving major Chinese firms uncertain whether domestic-only orders are even worth pursuing.
- Nvidia is openly frustrated, warning that export conditions have grown so burdensome that Chinese customers may walk away entirely — leaving the company with neither revenue nor the strategic leverage Jensen Huang promised Trump.
- Everything now pivots on a Trump-Xi meeting in the coming weeks, with Beijing still holding final approval authority over any imports and Nvidia reporting it currently earns zero data center revenue from China.
The Trump administration is quietly negotiating a compromise that satisfies almost no one. The proposal would cap Nvidia H200 chip sales to any single Chinese company at 75,000 units — a figure that sounds substantial until you learn that Alibaba and ByteDance have each privately told Nvidia they want more than twice that amount.
Reaching even this uneasy middle ground required months of back-and-forth. China rejected an earlier offer of the less-advanced H20 chip. Trump briefly considered allowing exports of the newer Blackwell line before senior advisors pushed back. The H200 — Nvidia's previous flagship, carrying six times the computational power of what had previously been permitted — emerged as the path of least resistance. Jensen Huang personally made the case to Trump that keeping Chinese AI firms dependent on American chips would blunt Huawei's ability to build a global competitor. Commerce Secretary Howard Lutnick echoed the argument publicly. Not everyone in the administration is convinced.
The constraints layered around the deal reveal how much internal friction remains. An overall ceiling of one million H200 chips has been set for China — more than many officials wanted, far less than Nvidia proposed. Exporters must certify their sales won't disadvantage American companies, and Chinese buyers must verify chips won't reach military users — a requirement made awkward by Washington's own designation of several major Chinese tech firms as military entities. The administration has also explicitly barred Chinese companies from deploying purchased chips in overseas data centers, closing off a commercially vital avenue that firms like Alibaba and Tencent had been counting on.
Nvidia has grown visibly frustrated, accusing officials of burying Trump's vision under bureaucratic conditions so onerous that Chinese firms may simply decline to buy. The company currently reports zero data center revenue from China and acknowledges it doesn't know whether Beijing will permit imports even if Washington approves them. The outcome now rests heavily on Trump's upcoming meeting with Xi Jinping — a conversation that could either unlock a fragile new trade arrangement or confirm that the needle was never there to thread.
The Trump administration is quietly working through the mechanics of a compromise that nobody seems entirely happy with. The proposal under discussion would cap how many of Nvidia's most powerful chips any single Chinese company could buy—75,000 H200 accelerators per firm, according to people briefed on the talks. It's a number that sounds large until you learn what the actual buyers want: Alibaba and ByteDance have each told Nvidia privately they'd like to purchase more than twice that amount.
The H200 is Nvidia's previous-generation flagship, the chip that powered the AI boom before the company released its newer Blackwell line last year. It has six times the computational muscle of what the Trump administration had previously allowed into China. Getting here required months of negotiation, rejection, and recalibration. China's government turned down an earlier offer of the less-advanced H20 chip. Trump considered allowing Blackwell exports but backed away after pushback from senior advisors. The H200 emerged as a middle path—powerful enough to matter, constrained enough to satisfy national security concerns.
The logic behind allowing any exports at all comes from Nvidia's leadership. Jensen Huang, the company's chief executive, convinced Trump that keeping Chinese AI companies dependent on American semiconductors would actually serve U.S. interests by preventing Huawei from building the revenue and developer base needed to compete globally. Commerce Secretary Howard Lutnick explained the argument on a podcast in January: such exports would foster "a positive economic relationship with China" and be "positive for us overall." Not everyone in the administration agrees. China hawks worry the chips will simply accelerate Chinese AI development with no American benefit. But this is, as Lutnick said, "the president's deal."
The constraints piling up around that deal reveal how much friction exists beneath the surface. The administration has set an overall ceiling of one million H200 chips total to China—more than many officials wanted, far less than Nvidia proposed. Individual company caps of 75,000 would allow a firm to build a midsize data center consuming roughly 100 megawatts of power. For context, OpenAI is planning facilities ten times larger as part of its $500 billion infrastructure initiative. Exporters must certify that sales won't reduce chip availability for American companies. They must also verify that Chinese customers will conduct rigorous due diligence to prevent military use—a requirement that creates obvious friction given that Washington has designated or plans to designate three of China's four largest tech companies as military entities, a classification those firms reject.
Another constraint has proven particularly limiting. Chinese companies wanted to use H200 chips in data centers overseas, particularly in Southeast Asia, to compete with American cloud providers. Alibaba submitted license requests for Malaysian deployments. Tencent's president noted in November that chip shortages force the company to prioritize internal use over renting capacity to external customers, constraining revenue growth. The Trump administration shut that door. Michael Kratsios, the White House's science and technology policy director, told lawmakers in mid-January that the rules explicitly prohibit Chinese companies from using these chips to build overseas data centers to compete with American hyperscalers. The impact on total purchase volumes remains unclear—none of the major Chinese firms have said whether they'll proceed with domestic-only orders.
Nvidia has grown visibly frustrated with the bureaucratic layering. The company accuses officials of undermining Trump's vision and argues that export conditions have become so burdensome that Chinese firms may simply decline to buy at all. The company didn't respond to requests for comment on the per-company cap proposal. AMD, whose MI325 chips would count toward the same customer limits, also declined to comment. Alibaba and ByteDance didn't respond outside business hours.
Much depends on what happens when Trump meets with Chinese leader Xi Jinping in the coming weeks. Trump said in December that Xi responded positively to the H200 offer, and Chinese regulators have instructed companies to prepare orders. But Beijing must ultimately approve any Nvidia sales into the country, and the government is balancing AI developers' hunger for cutting-edge American chips against years of effort to promote domestic alternatives from companies like Huawei. The administration's per-company caps would further complicate an already murky timeline. Nvidia reported last week that it still receives zero data center revenue from China and doesn't know whether Beijing will permit imports even if Washington approves them. The White House didn't respond to a request for comment on the upcoming meeting or the cap proposal.
Citas Notables
Keeping Chinese AI companies dependent on American technology will prevent Huawei from building the revenue and developer ecosystem necessary to compete globally.— Nvidia's argument, as explained by Commerce Secretary Howard Lutnick
When AI chips are in short supply, we prioritize internal use instead of renting it out externally. If there weren't an AI chip supply constraint, our cloud revenue would be growing more quickly.— Tencent Holdings President Martin Lau, November 2025
La Conversación del Hearth Otra perspectiva de la historia
Why does the Trump administration care how many chips go to any single Chinese company? Isn't a cap on total exports enough?
Because concentration matters for national security. If one company gets 500,000 chips, they could build a supercomputer that rivals anything in the U.S. Spreading the chips across multiple firms means no single entity gets powerful enough to pose what officials see as a military threat.
But Nvidia is arguing that letting Chinese companies buy these chips actually helps America. How does that logic work?
Nvidia's bet is that dependency creates leverage. If Alibaba and ByteDance are locked into American chips, Huawei can't build the ecosystem and revenue base to compete globally. It's a long-term containment strategy dressed up as commerce.
The article mentions China rejected the H20 chip but accepted the H200. What changed?
Timing and politics. The H20 was too weak for what Chinese companies actually needed. The H200 is powerful enough to be useful but still constrained compared to what Nvidia wanted to sell. It's a compromise that emerged after months of back-and-forth.
What about the overseas data center restriction? Why does that matter so much?
Chinese companies wanted to rent chips in Southeast Asia to compete with American cloud providers. Blocking that keeps them from expanding their global business. It's another way to limit their reach without explicitly banning the chips.
Does anyone actually think this will work—that China will use these chips only for civilian AI and not military purposes?
The administration requires certification of due diligence, but Lutnick basically punted when asked how he'd enforce it. He said Nvidia "must hew to" the license terms, then said he'd leave trust to the president. That's not a confident answer.