China is preparing to grant limited approvals for Nvidia’s advanced H200 chips to select leading artificial intelligence companies, including Alibaba, ByteDance, and DeepSeek. This move aims to ease their access to critical hardware while maintaining tight government control over the supply of these powerful processors.
The H200 GPU is a key asset for generative AI and high-performance computing, featuring 141 GB of HBM3e memory and a bandwidth of 4.8 TB/s, which significantly enhances training speed and model efficiency for large language models. Despite this demand, the anticipated approvals will total fewer than 200,000 units—well below what these companies originally requested—highlighting Beijing's cautious approach to allocating high-end AI chips.
China’s measured distribution of the H200 chips comes amid a broader strategic landscape. While the U.S. government authorized controlled exports of H200 and similar products to approved Chinese customers earlier this year, Beijing has held back on fully endorsing shipments. This means AI companies in China are navigating a complex environment where U.S. export permissions coexist with stringent Chinese controls.
Chinese firms are responding by accelerating efforts to develop domestic AI hardware. For example, DeepSeek is working on its own inference chips to reduce reliance on foreign technologies like Nvidia and Huawei. Other companies have shifted toward Huawei’s Ascend chips as a hedge against uncertainties in foreign supply.
Though this limited clearance signals progress for Nvidia, experts note it may not dramatically expand China’s AI capabilities but rather redistribute a constrained quantity of top-tier compute resources among its largest players. As a result, the country's AI sector continues to operate under a managed shortage, with select exemptions offering only partial relief from ongoing hardware constraints.