Nvidia was "stuck" with half a million chips. Then the solution was found

The AI chip market continues to surge, and China is attempting to maneuver between the desire to develop technological independence and the urgent need for advanced computing power: Regulators in China recently approved the entry of small quantities of H200 chips from American chip giant Nvidia into the country, according to a report in the Financial Times. According to the report, the move is intended to provide breathing room for leading technology groups in China that are competing head-to-head against American AI giants.

As part of the new approvals, technology giant ByteDance (operator of TikTok) and Tencent each received about 10,000 H200 processors in recent weeks. Additional Chinese technology companies are expected to receive similar approvals for shipments of this scope soon.

This is a chip that lags at least two generations behind Nvidia's most powerful processors, which Chinese customers are prohibited from purchasing as part of the US administration's strict export controls. Although under American licenses these Chinese companies were permitted to purchase up to 100,000 H200 chips each, the government in Beijing requires them to keep most of the equipment outside of China in order to continue encouraging the growth of local chip manufacturers.

As part of this interim solution, Chinese regulators made clear to the companies that they could transfer the H200 processors to Hong Kong - which operates as a separate territory outside China's customs border - and use them there, an option that is also covered by the American licenses. However, moving the chips to Hong Kong creates a significant hurdle: The former British colony currently lacks sufficient suitable data centers to store and operate them, and the power supply shortage in the area impedes rapid infrastructure expansion in the near term.

The restrictions and bans imposed by Beijing until now led to a widespread sales freeze, and Nvidia currently holds an inventory of about half a million H200 chips that were originally intended mainly for the Chinese market. In recent days, some of Nvidia's commercial partners, such as computing giant Lenovo - which integrates the chips with central processing units (CPUs) and communications equipment into full AI servers - have already notified their customers in China that orders for products based on H200 processors can be resumed. However, any actual purchase will still require specific and guided approval from China's National Development and Reform Commission.

The Chinese caution regarding chip imports reflects the administration's strategy to support local industry, led by Huawei, which aims to dominate the Chinese semiconductor market - the second largest in the world, to which Nvidia's access is mostly blocked due to American restrictions. On the other hand, China still suffers from a shortage of the most advanced chip manufacturing equipment, which limits its ability to satisfy surging domestic demand. Therefore, importing Nvidia chips remains necessary, particularly for the complex training stages of artificial intelligence models.

The recent regulatory relaxations on chip imports from the US were born out of concern about delaying local AI developers. Companies in China are attempting to produce an answer to advanced models in the West, including Anthropic's Mythos 5. Recently, the Chinese AI lab Moonshot presented its K3 model, which demonstrated performance very close to that of leading US models, and following it, Alibaba, DeepSeek, and Z.ai presented models at a similar level of capability, indicating a narrowing of gaps against American labs.

Currently, Chinese research labs are making growing use of locally produced chips for inference work - the action of generating real-time responses. However, for the much more complex and demanding training stage required to build knowledge infrastructures and identify patterns in models, most companies still rely on Nvidia's processors.