The United States has already given roughly a dozen Chinese companies the green light to buy Nvidia's second most powerful AI chip, the H200 - but not a single real delivery has taken place yet. Approved orders worth potentially billions of dollars thus remain "in limbo" just as CEO Jensen Huang is trying to break the stalemate in Beijing and find a way back into the market he used to dominate.
Huang eventually joined a delegation led by President Donald Trump en route to a summit with Chinese President Xi Jinping after Trump personally offered him a seat on board during a stopover in Alaska. From the market's perspective, hopes are pinned on the visit that it could unblock stalled H200 deals in China - but it also shows how deeply Nvidia is caught between US export policy and China's strategy of technological self-sufficiency.
Who is allowed H200 in China - and in what volume
According to leaked information, the U.S. Department of Commerce has approved about a dozen Chinese customers to purchase the H200 under license. On the list are the big names of Chinese Internet and cloud - Alibaba $BABA, Tencent $TCEHY, ByteDance or JD.com $JD - supplemented by a few distributors like Lenovo and Foxconn.
Each approved customer can take up to 75,000 units of the H200 chip under the licensing terms, either directly from Nvidia $NVDA or through approved distributors. In total, that's the potential for hundreds of thousands of accelerators, orders in the billions of dollars, which would represent a significant, albeit "truncated" recovery of Nvidia's market position, where it held an estimated 95% of the advanced AI chip segment before the tightening of export rules.
Before the tightening, US regulators estimated that China accounted for roughly 13% of Nvidia's revenue, with Huang himself talking about how the AI market there could reach $50 billion this year. All the more poignant is the fact that today he himself says Nvidia's share of AI accelerators in China has "effectively fallen to zero".
Why not a single shipment has taken place despite the approval
On a purely formalistic level, Nvidia and its Chinese customers have been given the green light by the US. But the reality is that the deals are being blocked primarily by Beijing. According to sources, after initial activity, Chinese firms began to back off after informal and formal "advice" came from Beijing to hold back.
Several factors are at play:
Self-sufficiency strategy: the Chinese government fears that another wave of H200 imports would weaken the push to build its own AI chip ecosystem. Companies like DeepSeek already publicly stress that they build their systems on homegrown accelerators, including chips from Huawei - and these examples are used internally as an argument that "it can be done without Nvidia".
New security regulations: the State Council recently issued two standards on supply chain security and launched a systematic review of dependence on foreign technology in critical infrastructure. This automatically puts foreign chips - including those from Nvidia - under more pressure.
Technical backdoor concerns: the specific structure of the agreed export regime means that chips must physically pass through US territory, raising concerns in Beijing about possible hardware or firmware interference.
The result is a stalemate: paper-approved contracts, effectively zero deliveries, and growing pressure within China for companies to learn to do without US accelerators.
Complicated conditions from the US side
US conditions are not easy either. At the beginning of the year, rules came into force requiring Chinese customers to prove they have "sufficient security procedures" and that the chips will not be used for military purposes. At the same time, Nvidia must confirm that it has sufficient chip stock in the US so that new exports do not threaten domestic supply.
Another layer has been added directly by President Trump: he has negotiated a structure whereby the US will receive 25% of the revenue from H200 sales to China. Since US law does not allow for the direct imposition of an "export levy" on a specific company, the workaround is that the chips must physically pass through US territory where the relevant levies can be assessed.
From an American perspective, this is the way to:
extract some of the economic value from sales to a strategic rival
while controlling the physical flow of chips
In Beijing, however, this reinforces mistrust - the combination of passage through the US and a high financial "tithe" is seen as both a technical (modification concerns) and political risk (leverage in future disputes).
Nvidia between two fires: export restrictions vs. Chinese self-sufficiency
The entire episode illustrates the delicate position Nvidia finds itself in. On the one hand, the US government is using it as a tool of its technology policy - export restrictions are meant to prevent China from catching up with the US in AI computing at an accelerated pace. On the other, the Chinese government is pushing domestic firms to reduce their dependence on Western hardware and help grow their own high-end players, led by Huawei.
Until two years ago, Nvidia controlled the vast majority of China's advanced AI chip market, and China was a double-digit part of its revenue. Today, Huang himself admits that the share of AI accelerators has "fallen to zero", and instead of horizontally "restricting" exports (i.e. banning a portion of customers), it is realistically in a situation where it is de facto blocked by both countries simultaneously.
Meanwhile, every additional month without H200 shipments increases the chances that Chinese players will switch permanently to domestic chips and Nvidia will lose a large part of the market there permanently. In terms of long-term strategy, it is therefore clear why Huang is pushing so hard for a breakthrough - even at the cost of a politically sensitive "co-travel" with Trump to Beijing.
What's next: scenarios for the H200 in China
The combination of US conditions, Chinese caution and political pressure on both sides suggests several scenarios:
Partial breakthrough: A compromise is negotiated, with limited H200 supplies under even stricter conditions (e.g. only through selected state or semi-state enterprises, with strong monitoring). Nvidia will get at least part of the sales back, but the fundamental change in the trend towards domestic chips in China will not be stopped.
Long term stalemate: Approved licenses will remain on paper, China will de facto block imports, and Nvidia will be replaced in AI infrastructure by a mix of domestic manufacturers and grey routes. H200 won't make it to China in any significant volume at all, Nvidia will focus on the US, Europe and other parts of the world.
The snag on one side: Further political escalation - new US restrictions or Chinese counter-steps - could make the matter a clear ban, which would lock the market in for good and send a signal to companies on both sides that the era of "globalized" AI supply chains is over.