US allows export of Nvidia H200 chips to China: what conditions are a game changer for AI and geopolitics
14.01.2026When Washington simultaneously sells the second most powerful AI chip in the world and tries to contain Beijing's technological breakthrough.
The US has officially opened the door to exporting one of the world's most powerful artificial intelligence chips, the Nvidia H200, to China, but has done so in such a way that Beijing's every move is tied to limits, checks, and financial "brakes."3 4 10 The Trump administration's decision lifts a nearly year-long de facto ban on the H200, while imposing a strict volume limit, third-party technical expertise and requiring Chinese customers to prove "adequate safety procedures."3 9 10 This compromise exposes Washington's main dilemma: how to remain the center of the global semiconductor industry without giving China the tools for military and technological breakthrough.
What Washington decided: H200 is possible, but not more than half of sales in the US
The US Department of Commerce, through the Bureau of Industry and Security (BIS), confirmed: Nvidia can sell H200 to Chinese customers, but the volume of chips sent to China cannot exceed 50% of the quantity sold in the US domestic market3 4 10 This limit is built directly into the licensing policy: each export request will be considered on a case-by-case basis instead of the previous “presumption of denial” introduced under the so-called AI Diffusion Rule in 2025.3 4 .
The new rule takes effect on Thursday and formally applies to the H200, Nvidia's second most powerful AI processor after the latest Blackwell line, which remains completely closed to China.3 4 10 Thus, Washington is trying to walk a “fine line”: give the market limited access to high-performance iron, but maintain an undeniable American lead at peak productivity levels.
What is Nvidia H200 and why is there so much politics surrounding it
The H200 is a Hopper family GPU, the successor to the H100, which has become the de facto standard for training large language models and AI data centers in the US and Europe.3 4 The H200 features increased HBM3e memory capacity and bandwidth, enabling significantly faster training of giant models and inference clusters compared to the previous generation.4 5 For Chinese cloud providers and internet giants, this is actually a ticket to the club of players capable of creating competitive global AI platforms.
This is why the H200 came under severe restrictions in 2025: in October-January, its export to China was actually banned, and Nvidia was forced to offer simplified versions like the H20, whose performance was artificially cut to the permitted thresholds.3 4 12 The H200 was seen as “too powerful” for the Chinese market, as it could overcome the restrictions of previous export regulations and accelerate the development of military and intelligence algorithms such as satellite imagery analysis, cyberattacks, or guidance systems.
How the new conditions work: limit, verification, security
According to TechWire Asia, Reuters and Bloomberg, the new regime for the H200 consists of several key restrictions1 3 10 :
- Volume limit: total exports of H200 to China cannot exceed 50% of the quantity sold on the US market during the relevant period.
- Technical verification: Every batch of H200s bound for China must be inspected by an independent laboratory headquartered in the US, which confirms technical specifications and the absence of unwanted modifications1 3 .
- Domestic market priority: Nvidia must demonstrate that "sufficient quantities" of H200 remain available for US customers and government projects1 10 .
- Requirements for customers: Chinese buyers must prove the presence of "appropriate security procedures" and sign a commitment not to use the chips in military and defense-related programs3 10 .
In fact, it is about creating a "controlled corridor" in which every H200 is visible, and every Chinese company becomes bound by conditions, the violation of which can lead to the instant closure of this channel.
25% in favor of the US: a political "AI tax"
A separate layer of discussion concerns money. Back in December 2025, Donald Trump announced on Truth Social that he would allow the sale of H200s to China “in exchange for 25%” — a quarter of the proceeds from these deals should go to the US budget.4 6 11 The publication FintechWeekly and a number of industry publications describe this as a radical update of the logic of export control: to the classic licenses, the mechanics of direct financial participation of the state in the circulation of technology are added.6 .
In the final version of the BIS regulation, the wording of “25%” is interpreted as an additional customs duty that is levied when the chips pass through the US, where they must be physically registered before re-export to China.4 6 Lawyers quoted by the specialized media note the atypicality of this model: export control is turning into not only a security tool, but also a source of income and an element of negotiations with Beijing.
How Nvidia lobbied for H200: meeting with Trump and the lost Chinese market
TechWire Asia and Bloomberg describe Nvidia's long lobbying marathon: after the H200 was subject to an almost complete export ban to China, the company effectively lost one of its key markets and risked pushing customers away to Chinese competitors.3 4 Jensen Huang, CEO of Nvidia, spent months convincing the White House and BIS officials that a complete blockade would only accelerate the development of Chinese alternatives and deprive the US of leverage over Beijing's technological trajectory.3 4 .
According to industry sources, Huang personally met with Trump in Washington, offering a compromise: allow the H200 in China with volume restrictions, technical verification, and financial benefits for the US.3 6 The new export regime appears to be the result of this bargain: Nvidia returns to the market, albeit not at full capacity, and Washington gains both control and revenue.
How China is reacting: from queues for chips to threats to block imports
Paradoxically, against the backdrop of the American “green light,” Beijing is playing its own game. According to Reuters, The Information, and Asian media, even before the formal US decision, Chinese regulators warned major tech companies (Alibaba, Tencent, ByteDance) that imports of H200 could only be allowed in certain cases — for example, for joint research projects with universities.5 7 10 Other companies were advised to scale back or postpone purchases until the government sets its own quotas and requirements.
At the same time, media outlets, citing Chinese sources, report that the Beijing authorities are considering a scenario of partially blocking H200 imports as a tool of pressure in negotiations with Washington.8 10 The idea is simple: reduce dependence on Nvidia, force Chinese players to rely on domestic GPUs, and demand broader concessions from the US in exchange for access to a huge market.
Chinese orders and shortage: 2 million H200s versus 700 in stock
Financial media outlets, including Yahoo Finance and Seeking Alpha, cite internal market estimates: Chinese technology companies have already generated preliminary demand for more than 2 million H200 chips, while Nvidia's actual available supply is currently estimated at around 700 units.2 10 Part of the production capacity of TSMC and related suppliers has already been reoriented to newer lines (Blackwell, future Rubin), so the H200 is more of a transitional product for Nvidia itself.
This creates a strange situation: Beijing theoretically gets the right to buy H200, but in practice faces a shortage and a queue, where priority is given to American cloud providers and the so-called “sovereign AI programs” of US-allied countries.1 4 12 Chinese players risk being “at the back of the queue,” even if they do not violate formal licensing restrictions.
Supply chain control: transit through the US and software beacons
As Tom's Hardware describes, the new regime involves not only paper licenses, but also technical checkpoints. Every H200 destined for China must physically pass through US territory, where the chips are checked by customs and related inspectors, creating a fully traceable logistics route.4 12 Only then can the GPU be re-exported to the Chinese end customer.
In addition, Nvidia offers customers software tools for geolocation verification: special software can record in which jurisdiction the chip is operating, which allows for additional tracking of possible attempts at re-export to third countries, such as Russia or Iran.4 12 For Washington, this is a way to reduce the risk of H200 “leaking” bypassing official channels, which was previously a serious gap in control over powerful GPUs.
Political controversy in the US: business vs. security
The new regime immediately became the subject of domestic criticism in the US. Some Republican and Democratic congressmen claim that the authorization of the H200, even with limits and duties, “strengthens China’s military and cyber capabilities” and undermines American superiority in AI, particularly in defense applications.3 6 10 They emphasize the difficulty of control: proving that none of the purchased H200s are used for dual purposes is practically impossible.
The Trump administration instead presents the solution as a pragmatic compromise: without licensed channels, China will still obtain powerful chips — either through a “gray” scheme or through its own developments — but then the US will lose both control and revenue.3 6 When the H200 is sold under license, Washington retains the ability to press the "stop" button at any time, use exports as a negotiating lever, and at the same time fuel its own semiconductor industry.
What this means for China’s AI giants
For Alibaba, Tencent, ByteDance and Baidu, the news about H200 is both a breath of fresh air and a cold shower. On the one hand, they get a chance to partially compensate for the "hole" after the H100/H200 ban, which hit their plans to create large language models that compete with Western ones5 7 On the other hand, they are stuck in a state of uncertainty: each contract must be coordinated with its own government, the BIS, and the US, and Beijing's position may change depending on its tactical goals in negotiations with Washington.5 8 .
This regime stimulates a double movement: big players continue to buy everything they can from Nvidia and, in parallel, invest in Chinese alternatives - from Huawei Ascend to domestic startups that receive government grants and benefits.5 8 In the short term, H200 will support the viability of their AI projects; in the long term, it will push for even more aggressive import substitution.
Why is this important for Ukraine and Europe?
For Ukraine, the H200 story is not just a story from the pages of the business media. It determines the pace at which China will be able to increase the computing power for its military, cyber troops and intelligence structures, which directly affects the global balance of power, including support for Russia.3 10 15 The longer the US maintains its technological advantage and control over the export of critical chips, the less chance Moscow has of gaining full access to advanced AI computing through Chinese channels.
Another aspect is important for the EU and, in particular, Ukrainian partners: the new regime of funds and duties shows that export control can become a source of financing for their own industrial policies.6 14 For Kyiv, which is building its own strategy for accessing AI hardware in times of war, this is a signal to look not only for direct grants, but also for models where joint projects with semiconductor manufacturers can be accompanied by flexible but controlled export schemes.
What's next for H200: the transition phase and the real battle for Blackwell
Despite the media focus on the H200, industry analysts remind us that this is already a transitional solution in Nvidia's product line. The company is actively launching the Blackwell architecture, and for 2026 an even more powerful Rubin line has been announced, which will further separate top systems from affordable Chinese chips.4 12 In this sense, the H200 release looks like a controlled “access to the previous generation,” while the real battle for technological superiority unfolds at the level of the next generations of GPUs.
At the same time, how the current export control model — with volumes, duties, and verification — works will determine the future design of rules for subsequent chips. If Washington finds the H200 formula successful, similar schemes could be applied to other manufacturers, including AMD and Intel, as industry sources have already hinted.6 14 This means that an “AI tax” and “controlled access to the past generation” could become the new global norm.
Sources
- TechWire Asia: "US Approves Nvidia H200 Chip Exports to China With Cap" — details about the 50% limit on US sales, BIS conditions, and Jensen Huang's meeting with Donald Trump.
- Yahoo Finance / video review: "US approves Nvidia H200 chip exports to China with some conditions" — confirmation of volume restrictions, security requirements for Chinese customers.
- South China Morning Post: “US government approves Nvidia H200 chip exports to China” — a description of the H200 as Nvidia's second most powerful AI chip, a description of the balance between containing China and maintaining market share.
- Tom's Hardware: "The Nvidia H200 export saga, as it happened" — the story of the year-long ban, the restart of exports, the requirement for transit through the US, optional geolocation control tools.
- FintechWeekly: "US–China Chip Tensions Renew Focus on AI Controls..." — analysis of the 25% duty model, comparison with classic export regimes, expectations for AMD and Intel.
- Bloomberg / Reuters (via aggregators): Previous reports of the Trump administration's intention to allow H200 in exchange for a 25% fee and volume restrictions.
- The Information / Times of India: materials on the position of Chinese regulators, conditions for R&D projects, Beijing's attempts to limit dependence on H200 imports.
- The News / other international media: reports of a possible partial blockade of H200 imports by China as an element of negotiating pressure.
- Seeking Alpha: "US formally clears Nvidia H200 AI chip exports to China..." — assessment of the shortage (2 million applications against 700 thousand available chips), market reaction.
- Semafor / Reuters (November 2025): First leaks about White House plans to soften the complete ban on H200 and move to licenses with control.
- Industry insights (Nvidia roadmap): context on the H100–H200–Blackwell–Rubin generations and their significance for the global AI computing market.
