Nvidia cleared to ship H200 AI chips to China under tight US terms

Allowing limited H200 shipments to China under Trump’s new revenue‑sharing regime underscores how difficult it is to fully decouple advanced AI compute while still wanting US firms to dominate the market. Nvidia is effectively threading a needle: using existing H200 inventory to serve a tightly controlled set of Chinese commercial customers, while assuring Washington that US buyers will not face shortages. From a capability standpoint, H200 is no longer Nvidia’s absolute frontier chip, but it is more than powerful enough to train and serve very competitive large models, especially with efficient architectures.

For the race to AGI, this keeps China’s top labs and platforms from being fully starved of cutting‑edge US silicon, even as they ramp up domestic accelerators and creative workarounds. It also shows how export control policy is shifting from simple on/off bans toward more intricate arrangements—caps by performance class, revenue skims, and vetting of end‑users—that try to slow military use without killing US corporate profits. In practice, that means global compute capacity for frontier‑scale systems continues to grow, but with more friction and geopolitical risk priced in.