China lets Nvidia's H200 chips trickle onto the mainland to help its AI firms keep pace with the US

China is allowing small batches of Nvidia's H200 chips onto the mainland to help domestic AI firms compete with the US, with ByteDance and Tencent each receiving around 10,000 H200 processors, according to the Financial Times. The H200 is at least two generations behind Nvidia's most powerful chips, which China cannot buy due to US export controls, and purchases require approval from China's NDRC planning agency.

China lets Nvidia's H200 chips trickle onto the mainland to help its AI firms keep pace with the US China is letting small batches of Nvidia's H200 chips onto the mainland to help domestic AI firms in the race with the US. According to the Financial Times, ByteDance and Tencent each received around 10,000 H200 processors, and more companies could follow soon. The H200 https://the-decoder.com/nvidia-unveils-new-h200-gpu-for-ai-and-hpc-workloads/ is at least two generations behind Nvidia's most powerful chips, which China can't buy because of US export controls. The US allows up to 100,000 H200s per company, but Beijing wants to support its own makers like Huawei https://the-decoder.com/china-reportedly-tightens-nvidia-h200-restrictions-limits-purchases-to-special-cases/ . Shipments to Hong Kong are also allowed, but the city lacks data centers and power. Purchases also need sign-off from the NDRC planning agency https://the-decoder.com/china-blocks-metas-2-billion-acquisition-of-ai-startup-manus/ . Chinese labs are catching up on the technical side. Moonshot with K3 https://the-decoder.com/moonshot-ai-releases-kimi-k3-open-weights-and-infrastructure-after-shaking-up-the-frontier-model-race/ , Alibaba with Qwen3.8 https://the-decoder.com/alibabas-qwen-team-releases-qwen-3-8-models-with-open-weights-under-the-apache-2-0-license/ , DeepSeek with DeepSeek V4 https://the-decoder.com/deepseek-launches-an-improved-v4-pro-model-raises-api-prices-and-makes-its-agent-software-open-source/ , and most recently Z.ai with GLM-5.3 https://the-decoder.com/zhipu-ai-releases-glm-5-3-claims-its-the-strongest-open-weights-coding-model/ . But they lag far behind US providers on available inference capacity. After a surge in demand, Moonshot had to turn customers away https://the-decoder.com/moonshot-pauses-new-kimi-k3-subscriptions-after-gpu-demand-maxes-out-in-48-hours/ . Nvidia keeps about half a million H200s in stock. AI News Without the Hype – Curated by Humans Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section. Subscribe now FT https://www.ft.com/content/6c5650fb-969d-4d4e-80d6-8d11002a8cf7