Chinese technology firms have bought more than 1 million Nvidia gaming and workstation chips this year, according to a Dealroom report, using consumer-class hardware to work around United States export controls on advanced AI accelerators.

The number matters because it measures the scale of a substitution trade. Washington restricts Nvidia's data-center products, such as the A100 and H100 families, from reaching Chinese customers. It does not restrict, at least not with the same force, the GeForce gaming cards and professional workstation GPUs that share a good deal of the same silicon DNA.

Buyers noticed. More than 1 million units have moved through this channel in a single year, Dealroom reports. That is a volume normally associated with retail gamers and design studios, not with companies assembling AI training clusters.

Why gaming chips can do AI work

The logic is straightforward. A modern gaming GPU contains thousands of CUDA cores, high-bandwidth memory, and the same software stack that Nvidia ships with its data-center parts. Software written for an H100 will, with modest effort, run on a stack of GeForce cards. It runs slower and less efficiently, but it runs.

For a Chinese firm that cannot legally import the top-tier part, a cluster of workstation boards is the second-best option. It is also, in some cases, the only option that arrives at all.

What the purchases reveal about the controls

The figure suggests the export-control regime has a wide side door. Rules drafted around specific product categories leave the consumer lineup largely untouched. Nvidia, for its part, designs its gaming and workstation lines for entirely different customers, and the volume reported here indicates how much demand has migrated into those lines from the restricted segment.

The pattern also carries a commercial signal. Chinese AI spending has not shrunk to fit the available legal supply. It has rerouted. One million chips is a rerouting of real scale, not a trickle of gray-market improvisation.

Who pays and who gains

The arrangement distributes benefits unevenly. Chinese buyers get compute, though at worse performance per dollar and per watt than the banned hardware would offer. Nvidia records revenue, though whether it can recognize and retain sales that end in sanctioned-adjacent use is a separate question regulators will eventually address.

The United States government, meanwhile, faces the familiar enforcement problem: controls on finished products are only as strong as the category definitions behind them. Each workaround tends to produce a new rule, and each new rule tends to produce another workaround.

What comes next

The obvious policy response is to broaden the definitions, pulling consumer and workstation GPUs into the licensing regime or capping their performance. The obvious market response, already visible in Beijing's push for domestic accelerators, is to treat every restriction as a procurement mandate for local silicon.

For now, the trade continues. Expect Washington to revisit the scope of its controls, and expect Chinese buyers to keep counting GeForce cards among their AI budgets until it does.

By Nathan Brooks