Sign in

US curbs open up to scrutiny key pathway Chinese AI labs use to obtain advanced compute: Expert

The expert noted Malaysia, Singapore, Thailand and Indonesia face curbs as they acted as transshipment networks

Published on: Nov 17, 2026, 14:08:27 IST
Share
Share via
  • facebook
  • twitter
  • linkedin
Copy link
  • copy link

The US Commerce Department on Monday issued an interim rule requiring licences for exports of advanced AI chips and servers to data centres anywhere in the world that may provide computing power to Chinese companies, and additional licensing requirements for Singapore, Malaysia, Thailand, Indonesia and Brazil. Ryan Fedasiuk, senior fellow at the Foundation on American Innovation, spoke to HT about what the measures mean for China’s AI ambitions, Washington’s partners in Southeast Asia, and for India. Edited excerpts:

Foundation on American Innovation senior fellow Ryan Fedasiuk.
Foundation on American Innovation senior fellow Ryan Fedasiuk.

How significant are the loopholes that Monday’s rule seeks to plug? How much would it affect China’s access to advanced compute for AI development?

This is a significant step in the right direction. It opens up to scrutiny a key pathway that many Chinese AI labs were using to obtain advanced compute. The rule allows the Commerce Department to require a licence before authorising the provision of certain amounts of computational power. Many members of Congress have long contemplated such a measure, for example in the Remote Access Security Act, and the executive branch is now undertaking it voluntarily. It is a major step in the right direction, and a long overdue one. But it is all going to come down to enforcement.

Deep Dive

Powered ByFAQAsk HT

There has been a lot of speculation that the Jensen Huang affair sparked this. What were the dynamics within the US government that led to these restrictions?

This is not the first time the US government has considered governing the provision of computational power. As early as the mid-2000s, the Commerce Department considered such a measure, before its legal office determined that the department did not have that authority. This rule is a major departure from that position: it asserts that the US government does, in fact, have the authority to require a licence for the provision of computational power located overseas. The Jensen Huang affair obviously had a major impact on the government’s willingness to take this step. You are clearly seeing multiple departments and agencies feel empowered to limit the provision of US technology to China in a way that maybe they hadn’t been before.

Why were five countries singled out? What does that say about where Washington thinks diversion is happening?

This is an initial step, and the purpose of the rule is to target a handful of countries where we know either transshipment or large-scale provision of remote access is happening. Southeast Asia has long been a primary transit hub for this kind of activity. But there are other markets not captured in this initial rule that the US government might want to look into down the line.

What are the specific concerns with the countries that have been named?

Malaysia, Singapore, Thailand and Indonesia face special restrictions, unsurprisingly, because they have acted as major hubs or transshipment networks for compute that is either bound for China or being rented by Chinese AI labs. Megaspeed is famously headquartered in Singapore. Thailand hosted GB300 servers that Moonshot used to train its Kimi K3 model, as the White House stated earlier this fall. Brazil is the outlier. It is not a market I track closely, to be honest, but it makes sense: it is a major hub of computing power that has sought to modernise its grid in recent years, and it has less stringent rules on disclosing who operates business entities. So it makes sense that the US government would have some concern about that market.

How is China likely to view this decision?

China is caught in a strange position. On the one hand, Chinese AI labs have clearly been exploiting this loophole for years to great effect. Recent research by a former OpenAI researcher, William Depeu, found that nearly every frontier Chinese AI model has in fact been trained on frontier compute, even when that is theoretically not supposed to be the case. So we know that closing this loophole will have a major effect on China’s AI industry. On the other hand, China cannot readily admit this, because it insists that no smuggling is happening and that Chinese AI labs are not dependent on foreign compute. That has been the thrust of Chinese policy for the past couple of years. So China should not retaliate. If it insists there was no loophole being exploited to begin with, there is a logical inconsistency in retaliating against a measure it says will have no material effect.

Does this make a material difference to the US-China AI race, and to China’s ability to develop the AI technologies it wants?

Yes, I think it has a significant impact. It is hard to assess exactly how much compute is available to Chinese AI labs through third-party vendors in Southeast Asia, but the short answer is that there is a lot of activity here. Singapore, for example, accounted for nearly 20% of Nvidia’s revenues in 2025, even though no major AI lab in the world is headquartered there. So you can assess, in broad strokes, that this kind of activity has represented a significant share of Southeast Asia’s AI buildout over the past few years.

India isn’t on the five-country list. Does that make India look like a more trusted destination for AI data centre investment than Southeast Asia?

India has been an important market for the United States, and it is a Pax Silica member. Obviously, there are issues in the trade relationship that both sides have been working through for the past couple of years. But I think the question for India is more about grid modernisation and the physical ability to host compute in the first place, rather than transshipment. At least, that is what we have observed so far. Of course, as India considers becoming a global compute hub and builds data centres of its own, it will be important for the United States to ensure that compute reaching India is not used to train restricted entities in China.

India is keen to attract global investment in data centres. How much does this rule add to the compliance burden for companies running data centre operations there?

This is an interesting question. In theory, the compliance burden from this rule should not be significant. Companies should already have some sense of who they are doing business with and who they are providing compute services to. It is a change, since this has not been required before, and it is a material shift in the way hyperscalers and neoclouds are accustomed to doing business. But realistically, it can be as simple as collecting basic know-your-customer information, and it should not necessarily impose a significant cost on data centre operators.

  • Shashank Mattoo
    ABOUT THE AUTHOR
    Shashank Mattoo

    Shashank Mattoo is the Washington DC-based US Correspondent for Hindustan Times. Shashank covers the Trump administration, the India-US relationship and America's policies towards South Asia and the broader Indo-Pacific.Read More

Get Latest News, breaking headlines, and top India News, and today real-time updates from across the nation. Stay informed on politics, government policies, crime, Today Weather, and major developments on Hindustan Times