A "trivial" number of Nvidia H200s reached China under US licence
Jeffrey Kessler, under secretary of commerce for industry and security, confirmed that a small number of H200 chips shipped to Chinese customers with approval. The licence regime works; the volumes say the market has moved on.
The word from the official is "trivial", and that is the finding. A licensing regime that approves shipments nobody takes up in volume is not restraining a market — it is documenting one that has already routed around it.
The demand-side explanation is the more interesting half. Alibaba, ByteDance and domestic developers have spent the restriction period optimising workloads for Huawei's Ascend architecture, and that work does not unwind because an export licence appears. Software written for a different accelerator is a sunk investment that keeps paying.
The evidence sits alongside the infrastructure. A gigawatt of Chinese AI capacity is running on domestic silicon with no NVIDIA inside it, and the lab operating it shipped a frontier model this week. Access to H200s is a smaller prize than it was when the controls were written.
Which is the awkward position for policy. The controls are being enforced, the loopholes are being closed, and the thing they were meant to prevent has been substituted around at a cost the substituting party was willing to pay. Success against the stated mechanism and failure against the stated goal are not the same measurement.
Bloomberg — Small Amount of Nvidia AI Chips Sold to China Via US License → · CNBC — U.S. trade official says 'very few' Nvidia H200 AI chips have been shipped to China → · CNBC — Chinese companies are ramping up homegrown AI chips, even if Nvidia is coming back →