Analysis
Nvidia's H200 chips are flowing into China again, in small shipments, according to reporting this week -- with major customers ByteDance and Tencent each receiving roughly 10,000 units in recent weeks. The shipments operate under a licensing framework the US government issued in January 2026, which allows each approved Chinese customer to purchase up to 75,000 H200 units.
The H200 is an older Nvidia architecture relative to the company's latest-generation chips, which remain off-limits to Chinese buyers under current export controls. Roughly ten Chinese firms have been cleared to purchase under the framework, including Alibaba, Tencent, ByteDance and JD.com, though actual shipments so far represent a small fraction -- reportedly around 13% -- of the legal ceiling.
“The H200 is an older Nvidia architecture relative to the company's latest-generation chips, which remain off-limits to Chinese buyers under current export controls.”
What's notable is which government is now the bottleneck: it's Beijing, not Washington, slowing the flow. China has pushed its major tech firms to keep purchased H200 hardware outside mainland China, supporting the growth of domestic chipmakers like Huawei and SMIC rather than deepening reliance on US silicon even where it's legally available. That's a reversal from the dynamic that dominated 2023-2025, when US export restrictions were the primary constraint on Chinese AI compute access.
The limited H200 flow intersects with this week's separate news that Samsung raised advanced chipmaking prices up to 15% -- both stories point to a global AI compute market where supply constraints, not just US-China policy, are shaping who gets access to what hardware and at what price. For US hyperscalers and AI labs, the H200's continued (if limited) availability to Chinese competitors is a reminder that export controls slow but don't fully sever Chinese access to Nvidia-class compute, even as China's own domestic chip industry works to reduce that dependency further.
The unresolved question is whether Beijing's reluctance to bring purchased H200s onshore reflects confidence in domestic alternatives closing the gap, or simply industrial policy aimed at forcing that gap closed faster than market incentives alone would achieve -- and whether either approach meaningfully changes China's AI training capacity in the near term.
