The US Commerce Department is finalizing a rule that would stop Chinese AI companies from remotely using advanced AI chips through overseas data centers. According to reports, the rule could be published as early as this month. It would extend the existing semiconductor export restrictions to seven countries, including Malaysia, Singapore and Thailand, with the aim of closing two channels at once: remote access and chip smuggling. Commerce is still consulting other government agencies and has not yet finalized the text.
As of publication, Commerce has not commented publicly, the rule text has not appeared in the Federal Register, and the full list of the seven countries has not been made public.
Targeting use, not just sales
Current chip export controls cover physical goods: high-end GPUs from Nvidia and others cannot be sold directly to Chinese customers, nor shipped to them through third countries. The loophole is that if the chips are legally installed in a data center in Singapore or Malaysia, a Chinese company can rent compute by the hour without a single chip ever entering China.
The new rule is aimed squarely at that kind of rental. The difference between the two approaches can be compared like this:
| Existing export controls | Proposed rule | |
|---|---|---|
| What is controlled | Export and transshipment of physical chips | Remote use of chips in overseas data centers |
| Enforcement lever | Export licenses, customs | Customer screening by cloud providers and data center operators |
| Role of Southeast Asia | Mostly transit points | All seven countries brought directly within scope |
The change in enforcement lever is the hardest part. Physical exports leave customs paperwork behind, whereas remote rentals depend on data centers verifying who the end user is, and there can be several layers of proxy companies in between. Until the details are published, it is unclear how Commerce intends to require data centers to carry out due diligence.
Legal authority and lobbying
Whether remote access counts as an "export" has long been disputed in the US. Export control lawyers have said publicly that, until Congress passes dedicated legislation, Commerce lacks clear legal authority over remote access. The US House of Representatives passed a bill on remote access earlier this year. Reports add that the White House had previously asked Congress to legislate so that Commerce could write this rule.
Nvidia has lobbied against closing this gap, arguing that US AI infrastructure needs to win in global markets. Jensen Huang stepped down as Nvidia CEO this month, and CFO Kress has become interim CEO; it is not yet clear whether the company's position on the new rule will change.
Southeast Asia has been a hotspot for data center construction in recent years, with a large number of new facilities concentrated around Johor in Malaysia. The US has already pursued multiple cases involving Nvidia chips shipped through third locations, and authorities in Taiwan have for the first time searched premises and seized servers over chip smuggling. If the new rule takes effect in the reported scope, operators in these countries will face a markedly heavier compliance burden.
For Chinese large-model companies, renting compute overseas has been a way to supplement training resources, especially when high-end chips are unavailable as physical hardware. With that route narrowing, the remaining options are to use domestic AI chips, rent compliant compute inside China, or find more savings in training methods. Which path costs more will only be clear once the rule text is out, with the list of seven countries and the transition arrangements.
Sources: The Information, The Defiant, CocoLoop, Tom's Hardware; the scope of the seven countries and the publication timing are as reported by the media, and Commerce has not yet released the rule text.