The headline says the US approved H200 chip sales to China. The more important detail is buried underneath: not one chip has actually been delivered, despite the approval being in place since December 2025.
Here is the backstory. Nvidia once controlled about 95% of China's market for advanced AI chips. Then, starting in 2022, the US began restricting what Nvidia could sell in China, and the restrictions tightened steadily over the following three years. China once made up 13% of Nvidia's total revenue. Today, Nvidia's CEO Jensen Huang has confirmed that Nvidia's share of China's AI chip market has effectively collapsed to zero.
In December 2025, President Trump approved sales of the H200 chip to vetted Chinese customers, with a 25% fee attached. The fee arrangement comes with an odd requirement: every chip sold to China must first physically pass through US territory, so that US customs can collect the money. Beijing immediately flagged this as a security concern, worrying the chips could be tampered with or fitted with hidden monitoring tools during that transit.
The US also added layers of paperwork. Chinese buyers must prove they have security procedures in place, certify the chips will not be used for military purposes, and Nvidia must hold enough inventory in the US before any China shipment is allowed. Getting through all of that on a case-by-case basis has moved slowly.
But the bureaucratic friction is only part of the story. The deeper issue is that China has decided, at the government level, that buying Nvidia chips right now works against its own industrial strategy. Beijing is pushing its biggest tech companies to buy domestic chips instead, particularly from Huawei. Huawei's chip revenue is projected to hit around $12 billion this year, up from $7.5 billion last year. ByteDance alone has reportedly committed over $5 billion to Huawei chip orders in 2026. The same ByteDance that is on the approved list to buy Nvidia's H200.
Huawei's latest chip, the Ascend 950PR, was launched commercially in early 2026 and outperforms the older H20 chips that Nvidia had previously been selling in China. More importantly, DeepSeek, which became the most talked-about AI model in the world at the start of this year, has already switched to Huawei chips and built tight compatibility with them. That kind of software integration matters enormously: when a country's most popular AI applications are designed to run on domestic chips, the case for buying foreign hardware weakens fast.
Some experts read China's reluctance as a negotiating tactic: hold out until the US offers access to something even more powerful, like the Blackwell chips, which remain fully blocked. That interpretation has some history behind it. China used similar hesitation over older, less powerful chips to eventually win access to the H200. The same playbook may be running again.
The deeper signal here is about leverage. Three years ago, export restrictions hurt China badly. Chinese companies scrambled for grey-market chips, engineers worked around hardware limitations, and AI development slowed. Today, China has closed enough of the gap, through Huawei's chips and through smarter software like DeepSeek, that the restriction is hurting Nvidia's revenues more than it is hurting China's AI progress. That shift in who needs whom is the most important thing happening in this story.
Nvidia entered the Beijing summit this week in a genuinely weak position. Its CEO was not even on the original delegation list; Trump reportedly called Huang personally and invited him mid-journey. The deal Nvidia wants, cleared shipments with stable Chinese demand, depends entirely on Beijing giving its companies the green light. And Beijing is in no rush to do that.