China is letting small shipments of Nvidia H200 GPUs through while still pretending it can live without US AI chips.
According to the Financial Times small batches of Nvidia’s H200 processors have entered mainland China as Beijing tries to help its top AI outfits chase US rivals. ByteDance and Tencent have each received about 10,000 H200 chips in recent weeks, according to two people familiar with the matter.
A few other Chinese tech groups could soon win approval for similar-sized shipments. The H200 is at least two generations behind Nvidia’s most powerful gear, which Chinese customers cannot buy because of US export controls.
Chinese companies have US clearance to buy up to 100,000 H200s each, but Beijing does not want most of them on the mainland. One person said regulators prefer that the chips remain outside China to keep pressure on buyers to use domestic silicon.
Regulators have told companies they can ship the processors to Hong Kong, which sits outside China’s customs border, and use them there. However, Hong Kong lacks the data centre space and power supply to house sufficient kit.
A person familiar with the mess said: “It’s a dilemma. Everyone needs the chips but struggles to find a way to use them in Hong Kong. The hope is for the control to loosen up gradually.”
Nvidia has about half a million H200 chips in stock, mainly for Chinese customers, but sales were jammed up by Beijing’s restrictions. Some partners, including Lenovo, told Chinese customers last week that they could resume ordering products containing H200 chips.
Those orders need separate approval through an application process overseen by China’s state planner, the National Development and Reform Commission. Beijing’s import controls are meant to give domestic chipmakers led by Huawei room to grow in the world’s second-largest semiconductor market.
China still lacks the most advanced chipmaking equipment, so it cannot meet local demand quickly enough. That means some Nvidia imports remain necessary, especially for training AI models rather than just running inference.
Chinese regulators have begun easing restrictions on US chips, allowing local AI developers to continue training cutting-edge models. Chinese labs are trying to compete with advanced systems such as Anthropic’s Mythos 5, while Moonshot, Alibaba, DeepSeek and Z.ai have released models suggesting the gap with US labs is narrowing.
Chinese outfits are increasingly using domestic chips for AI inference, where models generate live responses. Most still rely on Nvidia chips for the more complex training phase, leaving Beijing trying to starve the beast while feeding it.
