US to require licenses for overseas data centers renting AI chips to China

US to require licenses for overseas data centers renting AI chips to China

Under plans announced on Monday, the Trump administration will require overseas data centers that want to provide large amounts of computing power to China to obtain special licenses, and to deal with other new red tape. Until now, Chinese tech firms such as Alibaba and ByteDance could rent computing power from data centers abroad and so use powerful Nvidia chips they were generally barred from physically importing into China. That practice is known as remote access. The White House also imposed stricter rules on shipments of advanced chips to five countries, including Singapore and Malaysia, where US officials suspect significant levels of smuggling may have taken place. A White House spokesperson previously told WIRED the new rule would 'crack down on the smuggling of advanced US SI chips' and help ensure the most advanced chips go first to US innovators, not Chinese shell companies.

Will Knight, writing in Wired's AI Lab newsletter, says these are not new policy ideas; similar proposals have circulated in Washington for a while. But he finds the timing hard to ignore. Last week Nvidia cofounder Jensen Huang resigned as CEO after admitting he had an affair with Alice Huang, a Chinese businesswoman accused of having ties to Chinese intelligence services. She previously worked at Megaspeed, a Singapore-based data center company that offered remote access services and is suspected of chip smuggling. Knight speculates that the scandal may have been so embarrassing for the White House that officials felt compelled to show they had the situation under control. That link is his inference, not something the White House has stated.

Knight then reviews how the controls got here. The Biden administration unveiled the original rules in the fall of 2022, aiming to choke off China's access to the advanced chips needed to train the most powerful AI models, and to the equipment used to make them. It was a radical step: export controls had long been used to keep sensitive technology from specific military programs, governments and blacklisted companies, not to hold back the world's second-largest economy. The bet was that training top models needs huge computing power, and US firms like Nvidia dominate the chips that supply it.

Four years on, he writes, the US cannot claim total victory. Enforcement fell to the Bureau of Industry and Security (BIS), a relatively small, underfunded Commerce Department agency with a budget of just $235 million in 2026. Its job grew harder as successive rounds of complex restrictions were added and as a booming industry for smuggling AI chips into China took shape. At BIS, Megaspeed has become a poster child for the problem, though according to a person familiar with the matter it is far from the only suspected company, or even the largest. Bloomberg reported that roughly 50,000 of the estimated 136,000 AI chips Megaspeed imported into Malaysia and Indonesia in 2024 and 2025 were not accounted for in Nvidia's initial inventory checks. It is unclear where they ended up, and Nvidia later said it had found at least some of them. Even if all of them reached China, they would be a fraction of the chips believed smuggled in. Researchers at the Institute for Progress estimate China could have accumulated smuggled computing power equal to roughly 1 million Nvidia H100 chips by the end of 2026.

Chinese firms did not even need to smuggle. Renting overseas computing power did not violate US export controls, and policymakers and researchers knew it. Reasons it was tolerated: renting can be more expensive and cumbersome than owning chips, especially for long training runs; some analysts argued that letting Chinese firms use American technology abroad could keep them dependent on it rather than pushing them toward domestic alternatives; and restricting access would be hard to enforce and could hurt US technology companies. Now Washington is moving into that territory anyway. In 2022 it set out to stop Chinese companies acquiring certain American chips; now it is trying to regulate how they use those chips even when the hardware never enters China.

Whether that is wise depends partly on what the original restrictions achieved. Knight says there is evidence they made it considerably harder for Chinese AI companies to obtain computing power on a par with American rivals, but that assessing the precise impact has proven extremely difficult. The costs are real: New York Fed researchers studying earlier restrictions found affected American suppliers struggled to replace lost Chinese business, and estimate the restrictions wiped $130 billion from their combined market value while hurting profitability and employment. The controls also gave China a strong incentive to cut its reliance on American technology: Huawei developed a way to link powerful chips into larger megachip systems, and DeepSeek showed that more efficient training can yield competitive models with fewer resources. Knight closes by saying he worries the administration has overreached by trying to control how Chinese companies use American technology anywhere in the world.

Key facts

  • Announced Monday: overseas data centers that want to provide large amounts of computing power to China will need special licenses; remote access had been a legal route to Nvidia chips that China could not import.
  • The White House also tightened rules on advanced-chip shipments to five countries, including Singapore and Malaysia, where officials suspect significant smuggling.
  • Will Knight of Wired links the timing to Jensen Huang's resignation as Nvidia CEO after admitting an affair with Alice Huang, who once worked at Megaspeed; that link is his inference.
  • Bloomberg reported roughly 50,000 of an estimated 136,000 chips Megaspeed imported into Malaysia and Indonesia in 2024 and 2025 were not accounted for in Nvidia's initial checks; the Institute for Progress estimates China could hold smuggled compute equal to roughly 1 million H100s by end of 2026.
  • BIS, the enforcing agency, had a budget of $235 million in 2026; New York Fed researchers estimate earlier restrictions wiped $130 billion from US suppliers' combined market value.

Why it matters

Remote access was a legal way for Chinese firms such as Alibaba and ByteDance to use top Nvidia chips they could not import. Licensing the overseas data centers that serve them moves US policy from controlling which chips Chinese companies can buy to controlling how they use American technology anywhere in the world. Knight calls it another significant expansion of US efforts to shape China's technological development. The controls began in the fall of 2022 under the Biden administration, so this is the next step in a four-year escalation.

Who it affects

Overseas data center operators that sell large amounts of computing power to Chinese customers now face licensing and other red tape. Chinese AI companies that relied on rented capacity lose an easy route to advanced chips. Companies and officials in the five countries facing stricter shipment rules, including Singapore and Malaysia, are affected too. Nvidia and other US suppliers are exposed: Knight notes that restricting access could hurt US technology companies. The underfunded BIS gets the enforcement job.

How to use it

This is a policy story rather than a product, so there is nothing to adopt. Operators of data centers abroad serving Chinese clients should watch for the licensing details. The source gives no license criteria or effective date. Analysts can use the piece as a timeline of the controls, from the 2022 rules to the 2026 remote access plans.

How solid is it

The announcement and its main terms come from Wired's reporting. The Megaspeed chip figures are attributed to Bloomberg, and the 1 million H100-equivalent figure is an Institute for Progress estimate of what China could have accumulated, not a count. The claim that Megaspeed is a poster child but not the largest suspect comes from an unnamed person familiar with the matter. The link between the Huang affair and the new rules is the author's speculation, and the piece is a newsletter with opinion in it, including the closing worry about overreach. The source says assessing the controls' precise impact has proven extremely difficult.

Risks and caveats

Knight lists several risks: enforcement of remote access limits would be difficult, US companies could lose business (the New York Fed researchers put the loss from earlier restrictions at $130 billion in combined market value), and China has gained a strong incentive to build its own technology, as Huawei's chip-linking work and DeepSeek's efficient training show. He worries the administration has overreached. The source does not name the five countries beyond Singapore and Malaysia, and it gives no effective date, size threshold for 'large amounts of computing power', or license criteria. It does not say whether Alice Huang has been charged with anything, or who accused her.

“I worry that the Trump administration will soon find that it has gotten out over its own skis.”

— Will Knight, Wired AI Lab newsletter