Trump administration drafts rule to close China's remote chip loophole

The Trump administration is preparing a rule that would stop Chinese companies from renting American AI chips through data centers in other countries. 

This is a channel that the current US export controls do not cover currently.

Who is writing the rule and when will it land?

The Commerce Department’s Bureau of Industry and Security (BIS) is the office that has been working on the draft. BIS is also the office in charge of chip export licensing. 

People close to the matter say that the rule could get circulated to AI companies and other stakeholders as early as September.

Should the rule advance, it would replace the Biden-era AI Diffusion Rule, a framework that capped sales of advanced chips to certain countries. 

BIS scrapped that rule in 2025, saying it would have “stifled American innovation and saddled companies with burdensome new regulatory requirements.”

Why physical export controls miss the problem

While Washington’s export controls can block Nvidia’s top processors from shipping into China, it cannot stop a developer in Shanghai from paying to run workloads on those same chips sitting in a server rack abroad. 

Moonshot AI was accused of tapping GB300 systems in Thailand after it released its Kimi K3 model. Analysts say that the AI company did not do anything illegal, as it does not hold the hardware. 

Cassia King, a senior researcher on the Compute Policy team at the Institute for AI Policy and Strategy, stated that existing US policy “controls physical AI chips,” she said. “It does not cover remote access to those chips.” 

The White House, for its part, said that the administration “has implemented the most rigorous export control regime in modern history.”

How big the workaround has become

In November 2025, it was reported that Shanghai startup INF Tech was training AI on roughly 2,300 leading-edge Blackwell chips housed in an Indonesian data center, according to the Carnegie Endowment for International Peace. 

Indosat, the Indonesian telecom that bought the servers for about $100 million, said INF Tech only rents processing power and never touches the hardware.

Carnegie’s research also discovered that there were other firms doing the same. Alibaba and ByteDance have reportedly used Nvidia chips in Southeast Asian facilities to train their latest large language models.

Tencent struck a $1.2 billion deal last December for access to 15,000 B200 chips through a Japanese cloud provider. At least eleven state-linked Chinese entities have sought restricted US technology this way. The regional data center market that hosts much of this compute is projected to top $30 billion by 2030.

The legislative track running in parallel

The US Congress is also working to create legislation that covers the areas that the existing export controls do not cover. 

The revised Remote Access Security Act (RASA) has again cleared the House with bipartisan backing. 

The bill would redefine “exports” to cover remote compute access. In turn, it would force cloud providers to verify that their customers are not restricted parties. 

The decision to widen what falls under export controls is not welcomed by everyone. Oren Etzioni, founding CEO of the Allen Institute for Artificial Intelligence, is one of them, as he wrote in an op-ed that if RASA passes, foreign customers who currently buy American cloud compute “may instead begin patronizing China,” costing the US both revenue and long-term influence in the AI race.

If you're reading this, you’re already ahead. Stay there with our newsletter.