ByteDance and Tencent have each secured clearance to import roughly 10,000 Nvidia H200 AI processors directly into mainland China, as reported by the Financial Times. The shipments required explicit sign-off from China's National Development and Reform Commission (NDRC)—a pragmatic retreat from Beijing's aggressive import-substitution mandate in favor of keeping domestic AI labs competitive against US rivals.
While US export restrictions keep Nvidia’s cutting-edge Blackwell silicon strictly out of reach and leave Chinese labs working with two-generation-old architecture, engineering teams have no choice. Domestic developers are churning out frontier open-weight models, but local foundries and Huawei's Ascend hardware cannot bridge the acute inference bottleneck. Without sufficient compute throughput, even top-tier Chinese AI startups have recently been forced to turn enterprise users away during traffic surges.
Workarounds through offshore hubs have hit a hard physical wall. Hong Kong still permits hardware imports, but the territory’s data center footprint and power grids are strained to capacity. With Nvidia holding an estimated 500,000 H200 units ready to move, Beijing’s state-backed self-reliance campaign must settle for a compromise: importing older American silicon under strict regulatory caps to prevent its premier tech giants from falling fatally behind.