The Trump administration is drafting legislation to stop China from renting advanced GPU computing power through cloud services based in third countries such as Vietnam and Singapore. The move extends export controls beyond physical chips to the remote use of U.S.-made hardware, a gap that has let Chinese firms train AI models without ever importing a single banned processor.
Physical chip export controls have proven difficult to enforce. Thousands of banned GPUs get smuggled into China each year with switched labels and deceptive packaging, a problem I reported on last week. Now U.S. officials are targeting a different vector: the data center itself.
The Remote Access Security Act
The Remote Access Security Act, introduced last December, is the centerpiece of this effort. Under current law, bad actors can train AI models by accessing advanced chips under U.S. jurisdiction without a license from the Bureau of Industry and Security. That agency currently lacks authority to require licenses for remote access, and the new legislation would change that.
Senator Dave McCormick, the bill's sponsor, framed the issue in stark terms. "This is a critical security gap that allows our adversaries to use our most advanced technology against us," he said. "We must close it before they close the gap on us."
The bill would require U.S. cloud providers to verify user identities and block access from entities linked to China's military or AI development programs. It is designed to cover not just domestic data centers but also U.S.-controlled facilities abroad, including those in Vietnam and Singapore that have become popular hubs for Chinese AI firms.
A House Plan With Bipartisan Support
The House has a plan to implement another restrictive piece of legislation, with sponsors from both parties. Representative John Moolenaar (R-MI) said the goal is to end what he called China's backdoor access to American computing power. "China has found a way around our export controls by simply renting what they cannot buy," Moolenaar said. "This legislation shuts that door."
Representative Josh Gottheimer (D-NJ), a co-sponsor, echoed that view. "Our adversaries are renting what they can't buy," he said. "That's a loophole we cannot afford to leave open."
The House version would also require U.S. cloud platforms to verify user identities to prevent China's AI buildup. It mirrors the Senate bill in key respects but includes additional reporting requirements for cloud providers that serve foreign clients.
The Messy History of Nvidia Export Controls
Nvidia export bans have been inconsistently applied. The H20 chip, designed specifically for the Chinese market, was allowed, then banned, then allowed again. Jensen Huang, Nvidia's CEO, has historically influenced these decisions by meeting directly with Trump, and his lobbying has led to sudden reversals in policy.
The U.S. sometimes caps GPU export numbers to firms like Alibaba, limiting how many advanced processors a single company can acquire. But those caps apply to physical shipments, not to remote access. A Chinese firm can simply rent time on a U.S.-based GPU cluster in Singapore, train its models, and never touch a physical chip.
That is the loophole the new legislation targets. The Bureau of Industry and Security would gain authority to license remote access under the proposed rules, closing the gap that currently lets foreign actors use U.S. computing power without oversight.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Attrition, Not a Total Ban
The strategy aims to slow China's technological advancement through attrition, not a complete ban. The article questions whether the U.S. can police another country's access to remote services, and suggests that blocking cloud services is even harder than controlling hardware. Bits cross borders more easily than chips, as GPT put it in a recent analysis.
That observation cuts to the heart of the problem. A physical chip can be stopped at a port. A remote connection cannot be stopped at a border. The legislation would require U.S. cloud providers to police their own networks, but enforcement would depend on cooperation from foreign governments and private companies.
The article posits that the U.S. may be trying to slow China down by attrition, like the Cold War. The goal is not to stop Chinese AI development entirely but to make it slower, more expensive, and more difficult. Every month of delay buys the U.S. more time to maintain its lead.
A Global Race, Not Just a Bilateral One
Other nations are also racing to create national-language and domestically controlled AI platforms. The Sino-American race is of global interest, and countries from Europe to Southeast Asia are watching closely to see which model prevails.
The ultimate impact on China's AI progress remains uncertain. The article asks if the legislation is just saber-rattling or can actually hobble China. The answer depends on enforcement, international cooperation, and China's ability to find alternative sources of computing power.
The Remote Access Security Act and the House version are still working their way through Congress. If passed, they would represent a significant expansion of U.S. export control authority, moving beyond hardware to the services that make hardware useful.
The author, John Werner, is an MIT Senior Fellow & Lecturer, 5x-founder & VC investing in AI. He previously wrote extensively about export controls for Nvidia chips and reported last week on GPU smuggling. His analysis suggests that the U.S. is entering a new phase of the technology competition, one where the battlefield is not just ports and factories but the cloud itself.
For now, the legislation faces an uncertain path. Enforcement would require U.S. cloud providers to track users across borders, a technical and legal challenge. And China has shown a remarkable ability to adapt to U.S. restrictions, whether through smuggling, alternative suppliers, or domestic chip development.
The year 2026 will likely determine whether this approach works. If the legislation passes and is enforced, it could slow China's AI progress. If it fails, the U.S. will have to find another way to maintain its edge.
The stakes are high, and the clock is ticking.