The Trump administration is drafting a new AI export control designed to close one of the biggest remaining loopholes in US chip policy: China's access to advanced AI computing power through remote servers located in nearby countries, The Information reported on August 28. According to a report by Tom's Hardware, the rule would specifically target remote access routed through Thailand and Singapore, two hubs that are not subject to the same export restrictions as China, and the Commerce Department could share a draft with trade groups for feedback as early as September.
The move marks the next chapter in a policy fight over breaking AI news that has repeatedly reshaped the global AI hardware market. Since the Biden-era AI Diffusion Rule was shelved in early 2025, the United States has struggled to settle on a coherent framework for keeping advanced AI compute out of Chinese hands — and Beijing's labs have adapted faster than Washington's rulemaking.
The loophole: compute without borders
Traditional export controls govern the physical shipment of chips. But AI models do not need chips to be physically present in China. A researcher in Beijing can rent GPU time from a data center in Thailand or Singapore, train or run models remotely, and never take possession of restricted hardware. That is the gap the proposed rule is meant to close.
If the Commerce Department moves forward, it would be attempting something no US agency has done at scale: regulating not the export of goods, but the export of compute itself.
The Kimi K3 catalyst
According to The Information, one of the driving forces behind the proposed rule is Moonshot AI's Kimi K3 model. Michael Kratsios, director of the White House Office of Science and Technology Policy, alleged in a July post on X that Moonshot used distilled US models to train Kimi K3 through Nvidia-equipped servers in Thailand.
"We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model," Kratsios wrote, describing what he called a sophisticated internal platform for large-scale distillation against US models. The allegation has not been tested in court, and Moonshot has not publicly responded to it in detail, but it has become the reference point for officials arguing that remote compute access is already being used to sidestep US restrictions.
Legal doubts shadow the plan
Even administration allies concede the rule faces steep legal hurdles. An attorney at Baker McKenzie told The Information that it is "widely acknowledged" the Commerce Department cannot enforce a regulation on remote access, because the department has traditionally regulated the transportation of physical goods rather than intangible services.
The Commerce Department could still pursue the same goal through other means, notably know-your-customer checks on cloud providers. The Biden administration put similar requirements into effect with its Foundry Due Diligence Rule in early 2025, but the Trump administration has said it will not enforce that rule — leaving the policy landscape a patchwork of enforced and abandoned authorities.
A rulemaking history of reversals
The uncertainty is not accidental; it reflects a chaotic two years of AI trade policy:
- Early 2025: The Biden-era AI Diffusion Rule, which would have imposed a tiered global licensing structure for advanced AI chips, was overturned before taking full effect.
- March 2026: The Commerce Department issued a statement about a new tiered licensing structure, then withdrew it roughly a week later after pushback from the US AI industry.
- August 2026: The new remote-access rule emerges, with a draft possibly shared with industry as soon as September.
Each reversal has left data center operators, cloud providers and chipmakers guessing which rules will actually be enforced — and building compliance strategies for regimes that may not survive the month.
Enforcement is moving anyway
While the remote-access rule remains a draft, enforcement around the physical supply chain has accelerated. Taiwan detained an Nvidia employee in July on suspicion of falsifying documents tied to chip shipments, and on August 24 Taiwanese prosecutors indicted nine people — including a manager at Nvidia and two former Supermicro employees — over an alleged scheme that smuggled 74 AI servers containing restricted Nvidia chips into China.
For Beijing, the proposed rule would compound an already tightening squeeze. Chinese firms have been working around US curbs with domestic accelerators, stockpiled hardware and now, allegedly, rented foreign compute. Cutting off the rental channel would leave fewer workarounds — and push Chinese labs to invest even more heavily in homegrown silicon.
What happens next
If the Commerce Department shares the draft with trade groups in September as reported, expect a fierce comment period. US cloud providers with Southeast Asian data centers will likely argue the rule is unenforceable and harmful to legitimate business, while security hawks will press for the strictest possible know-your-customer regime. Legal challenges are all but guaranteed either way.
What is clear is that the battleground of AI export policy has shifted from the shipping dock to the cloud console. The country that writes the rules for remote compute access — and makes them stick — will shape how advanced AI capability diffuses around the world for years to come.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
