Nvidia's export-restricted AI chips are trading in China at roughly twice their official US price, and Washington still considers its export control regime effective, according to a Startup Fortune report.

The gray market pricing marks the clearest quantitative measure yet of how much distance separates US policy intent from on-the-ground reality in China's AI hardware supply chain. Restricted Nvidia GPUs — the silicon at the center of successive rounds of US export rules — continue to reach Chinese buyers through intermediaries, resellers and circumvention channels, with the doubling of street price functioning as a direct tax on restricted access rather than an effective blockade.

The pricing dynamic tells its own story. When a chip sells for 100% above its list price inside the restricted market, demand has not collapsed — it has simply repriced. Chinese AI developers, cloud operators and research groups are still acquiring the compute they need, paying a premium that reflects smuggling costs, intermediary margins and the legal risk absorbed by middlemen. The buyer who pays double is a buyer who still gets the silicon.

Washington's position, as characterized in the report, is that the controls are working. That assessment rests on the argument that restrictions raise costs, slow procurement at scale and deny Chinese customers the volume access they would otherwise enjoy. There is some truth in the volume argument: gray market channels move chips in ones, tens and hundreds, not the tens of thousands that a hyperscaler or a national AI champion would order directly from Nvidia through legitimate distribution.

But the premium pricing cuts both ways analytically. It demonstrates persistent, inelastic demand for Nvidia's restricted product families among Chinese customers. It also shows the enforcement perimeter remains porous enough that supply, however throttled, continues to flow. A control regime that doubles the price of a component without eliminating availability functions as friction, not as a wall — and friction is something well-capitalized buyers can pay through.

The commercial picture for Nvidia itself is mixed. The company cannot legally sell restricted GPUs into China, which cuts it off from a market that was once worth billions in annual revenue. Yet chips bearing its name still clear at doubled prices in Shenzhen-style trading channels, evidence that its products remain the default choice for Chinese AI work even under sanctions. No domestic Chinese accelerator has displaced that preference at the premium end, despite Beijing's sustained push behind domestic silicon alternatives.

For US policymakers, the finding sharpens an uncomfortable question about measurement. If the metric of success is price inflation and inconvenience for Chinese buyers, the controls deliver. If the metric is denial of advanced compute to Chinese AI programs, the doubled-price gray market suggests a substantial gap between the two. The report's framing — chips selling for twice their price while officials affirm the policy is working — captures that gap in a single image.

The enforcement burden falls on tracking individual hardware units through global distribution, a task made harder by the chips' small physical footprint and the depth of the global electronics intermediation chain. Each round of US rules has tightened license requirements and lowered the performance threshold for restricted parts, and each round has been followed by reports of continued availability inside China at escalating markups.

The pattern sets up a predictable competitive dynamic going forward: expect continued tightening of US rules alongside continued gray-market arbitrage, with Chinese buyers paying escalating premiums for restricted Nvidia silicon until a domestic alternative — or a further escalation of enforcement — changes the economics on the ground.