Economy & Policy

Commerce closes Nvidia remote-access loophole, targeting Singapore

Commerce issued its first export-control rewrite since the Nvidia scandal, requiring licenses to ship advanced AI chips to data centers in Singapore, Malaysia, Indonesia, Thailand and Brazil, and closing the remote-access loophole used by Chinese firms.

By Olivia Hart

7 min read

The Nvidia scandal fuels tougher China chip rules as Commerce targets Singapore and Malaysia
The Nvidia scandal fuels tougher China chip rules as Commerce targets Singapore and MalaysiaPexels / pixabay

What's News

  • Commerce issued an interim final rule on Monday, its first update to the export control regime since the Nvidia scandal broke
  • Licenses required to export advanced AI chips to data centers in Indonesia, Malaysia, Singapore, Thailand and Brazil, plus any remote access by China-headquartered companies anywhere
  • Existing data centers serving Chinese-headquartered companies have 90 days from the effective date to obtain a license
  • Roughly 600,000 Nvidia H200 chips sit idle in Hong Kong because China has withheld import approval, Reuters reported
  • Four bills are in this year's National Defense Authorization Act, including the MATCH Act, which gives Japan and the Netherlands 150 days to align controls with Washington

The U.S. Department of Commerce issued its first rewrite of chip export controls since the Nvidia scandal broke, issuing an interim final rule on Monday that requires licenses to export advanced AI chips to data centers in Indonesia, Malaysia, Singapore, Thailand, and Brazil. The rule takes effect immediately once published in the Federal Register.

The measure closes a long-standing loophole that let Chinese companies reach AI chips sitting in overseas data centers, and it applies the licensing requirement wherever a large data center operator worldwide wants to grant access to a company headquartered in China. Existing data centers serving such companies get 90 days from the effective date to obtain a license.

The backdrop is the resignation of Nvidia CEO Jensen Huang on Thursday, which removed the loudest industry defender of U.S. restrictions on advanced chips and chipmaking equipment sold to China.

Huang had argued such restrictions would spur the creation of a competing Chinese chip industry and that it was better for the U.S. to keep Chinese companies hooked on American technology. He also backed open-source AI models — many developed by Chinese firms such as DeepSeek — against claims they posed serious cybersecurity risks.

Revelations that Huang had an affair with Alice Huang, a former executive at Megaspeed, have damaged his credibility on export controls. The source did not say whether Jensen Huang's resignation was linked to the affair.

What does the new rule actually change?

Current controls already block the most advanced AI processors from Nvidia, AMD, and Intel from being sold to Chinese companies, though the Trump administration has allowed sales of Nvidia's H200 to approved Chinese customers. Chipmaking equipment and high-bandwidth memory are also blocked. The new rule adds geography, not categories: licenses become mandatory for advanced chips bound for data centers in five named countries and for remote access from anywhere.

Megaspeed, the Singapore-based company accused of funnelling Nvidia chips to China, has denied sending chips to China. Malaysia was reportedly home to Megaspeed's data centers.

The administration itself may be split on how to attack chip smuggling. The Commerce Department press release notes the rules were "written by and at the direction of the Department of Treasury and the National Security Council." Congress may act if the executive branch stays divided.

Can Congress turn these rules into law?

Several bills sit in this year's National Defense Authorization Act, and experts say the scandal has raised the odds of passage.

"The regulatory path in Washington is going to be harder and bumpier," Hanna Dohmen, a senior research analyst at Georgetown's Center for Security and Emerging Technology, told Fortune before the rule's release. "The Huang scandal adds momentum to push those bills through, and the chance of those passing are likelier than ever."

Joshua Kurlantzick, a senior fellow at the Council on Foreign Relations, put it more plainly: "No matter what Trump does, it's possible that Congress should be able to provide limitations. It's not only about Trump anymore."

  • Remote Access Security Act: Seeks to close the remote-access loophole, giving Commerce authority to license and penalize foreign companies tapping advanced U.S. AI chips from abroad. The House passed it in January; the Senate has not yet acted.
  • Chip Security Act: Requires advanced chips to ship with security measures such as location tracking so processors are not diverted on export, and requires exporters to promptly disclose evidence of diversion or smuggling. It stops short of mandating kill switches.
  • AI Overwatch Act: Treats advanced chips like weapons — before any sale to China or other adversaries, the administration must notify Congress, which could block the deal with a joint resolution of disapproval. It also carries an "America First AI" provision requiring Commerce to certify that exports will not leave U.S. customers short of computing power.
  • MATCH Act: The most aggressive of the four. It would largely ban the sale and servicing of deep ultraviolet lithography tools across China and impose broader restrictions on facilities tied to major Chinese chipmakers such as SMIC and CXMT. Allies including Japan and the Netherlands get 150 days to align their own controls with Washington's; if they fail, the U.S. can extend controls to equipment supplied from those countries even when it contains no U.S.-made components.

Rules on paper, however, do not move chips. "Knowing that a chip was smuggled matters little if the Bureau of Industry and Security doesn't act," Dohmen said. Enforcement is the gap the legislation cannot close.

How might Singapore and Malaysia respond?

"The administration wants to encourage Malaysia and Singapore to remain in the American AI ecosystem, but it doesn't want them to become conduits through which Chinese firms can obtain advanced chips — or remotely access the computing power those chips provide," Jennifer Lind, associate professor of government at Dartmouth College, said. "That has been its policy so far and I don't anticipate this changing."

The new rule includes a carve-out for end users that are, or are owned by, a U.S.-headquartered company. The region has attracted billions of dollars from U.S. tech companies such as Microsoft, Oracle, and Amazon, which should qualify for that exemption, though the remote-access limits still bind them.

On enforcement, Kurlantzick is "more inclined" to believe Singapore is taking smuggling seriously. Malaysia, in contrast, "wasn't really doing anything, and didn't care," he said. "The U.S. is insanely unpopular in Malaysia."

For Malaysia, the controls may cut the wrong way. The country wants to become a data center hub, and Kurlantzick said "For Malaysia, the controls could be disastrous. From what Prime Minister Anwar Ibrahim sees, their biggest potential future industry is a data center and AI hub for Chinese companies."

Washington is also pressing the region into its "Pax Silica" framework, an initiative to secure semiconductor and critical mineral supply chains that bypass China. Beijing counters with the World Artificial Intelligence Cooperation Organization, and the U.S. wants countries to pick one side. Trump is in Manila this week for the ASEAN Summit, where he is expected to lobby Southeast Asian countries to join Pax Silica — and answer awkward questions about the expanded rules.

Kurlantzick doubts the appeal. "The U.S. isn't really offering Southeast Asian countries — or really anyone in Asia — much in return for their demands."

Does China still care about export controls?

The biggest barrier to chips entering China right now is Beijing, not Washington. The Chinese government has largely withheld permission for its companies to import Nvidia chips, pushing them instead toward domestic chips from Huawei. DeepSeek has led the effort to run Chinese AI models well on Huawei's Ascend processors.

"Export controls still matter," Lind said, though she conceded they may matter less. "China is already well on its way toward developing its own AI ecosystem; Huawei makes its own chips for training and running AI models, and a software platform (CANN) that's an alternative to Nvidia's CUDA ecosystem."

What she watches now is "whether semiconductor manufacturing equipment continues to be restricted — whether China develops the capability to manufacture its own frontier chips," she said. "The Trump administration continues to restrict SME exports, and will likely continue to."

Huawei's Ascend processors still trail Nvidia's best chips, particularly for training models.

There is also an inventory angle. Roughly 600,000 H200 chips are gathering dust in Hong Kong, Reuters reported, because China has withheld import approval. Before the BIS rule was announced, Dohmen asked: "Would the Chinese government be more willing to let those H200s flow into China? If there is certain significant demand for remote access, and that's cut off, there may actually be more need for the H200s."

The constraint is supply, not just policy. China "does not have the capacity at this point to meet that demand," Dohmen said.

The next test arrives in Manila this week, where allies will be asked to sign up for Pax Silica while their data centers absorb licensing costs and Beijing weighs whether to release those 600,000 stranded H200s.

Source: fortune.com

Share this article:

More from Olivia Hart

Olivia Hart

Show full bio

Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,205 articles

Related articles

« Previous articleNext article »