The United States has spent years blocking Nvidia's most powerful AI chips from reaching China. Now Washington is looking at the workaround that made those rules leakier than intended: the cloud.
According to CNBC, Chinese AI firms have reportedly tapped advanced Nvidia computing power located overseas — renting access to the hardware rather than importing it — a practice that tests the limits of U.S. export controls. Lawmakers are weighing whether to close that cloud-access gap, CNBC reports.
The timing is awkward. Tom's Hardware reports that China has approved its first Nvidia H200 deliveries to ByteDance and Tencent under case-by-case import licenses. But the approvals come with a catch: most of each company's U.S.-licensed allowance — understood to be up to 100,000 units apiece — must stay outside the mainland. In other words, the chips are cleared, but largely on the condition that they sit somewhere else, which is precisely the arrangement that has raised concerns in Washington.
Tom's Hardware also frames the opening as arriving too late for Nvidia, arguing that homemade Chinese chips have already cornered the domestic market as the country pushes toward semiconductor independence.
That is the tension at the center of this story. Export controls were designed around physical shipments — boxes crossing borders. Modern AI computing does not require the box to move; a data center abroad and a network connection can do the job. If lawmakers act, the rules would extend from what China can buy to what China can log into.
Why it matters: how Washington resolves this determines whether U.S. export controls still meaningfully limit China's AI ambitions, or whether they mostly cost American chipmakers a market that is already slipping away.