Chinese AI firms seek overseas Nvidia computing capacity over fears of U.S. curbs

Chinese AI companies are trying to secure Nvidia-powered computing capacity outside China as they fear Washington could tighten access to advanced processors. Reuters reported on November 12, citing three people familiar with the matter, that firms were exploring overseas rentals—a route that lets them use chips they cannot import directly.

A feared policy response prompts capacity inquiries

The concern, according to Reuters’ sources, was that the scandal involving Jensen Huang could prompt tighter U.S. restrictions on remote use of advanced processors. They linked the capacity search to that possible policy response.

A major Chinese cloud provider had discussed internally whether to lock in Nvidia computing capacity overseas. Some companies had contacted data-center operators about reserving it. These were discussions and inquiries, not confirmed capacity reservations.

The people who spoke to Reuters declined to identify themselves or name the cloud provider because the matter was sensitive.

Renting the computing power, not importing the chips

Washington restricts exports to China of Nvidia’s most advanced AI chips. Chinese companies can nevertheless access those processors remotely by renting computing power at overseas data centers. The hardware stays abroad while customers pay to use its computing capacity.

A reported licensing rule would change the terms

Gizmodo, citing The Information’s November 11 reporting, said the Commerce Department was finalizing a rule intended to make remote access harder for Chinese AI developers. The measure was described as being finalized, not as an enacted restriction. Its reported structure combines licensing, disclosures and a conditional exemption:

• Licenses for chip deliveries: Local cloud providers in Singapore, Malaysia, Thailand, Brazil, Indonesia, Norway and South Africa would need U.S. government permission before receiving Nvidia Blackwell chips or more advanced models.
• Recipient and use disclosures: Companies would have to submit reports to Commerce identifying who would receive the chips and what they would be used for.
• A conditional exemption: U.S.-operated data centers would avoid the licensing requirement if they blocked remote access for customers in restricted countries such as China.

Nvidia has argued against that approach. Gizmodo said the company had opposed the rule through lobbying, warning that blocking remote access could push China to accelerate its own chipmaking efforts and potentially gain an advantage in AI. That is Nvidia’s forecast, not an established outcome.