The Trump administration is drafting a new rule aimed at stopping Chinese firms from renting access to advanced AI chips they aren’t allowed to buy, according to The Information. The move targets a well-known gap in US export controls: even when China can’t purchase Nvidia’s top processors outright, its companies can still tap that computing power remotely through overseas cloud providers. The Information reports the administration is working on regulation to close that loophole.
This is significant because it shifts the battleground. Until now, US policy focused on the physical chip. You couldn’t ship an H100 to Shenzhen. But you could spin up a cluster of those same chips in a data center in Singapore, Malaysia, or the Gulf, and let a Chinese customer log in and train a model on them. The hardware never crosses the border. The compute does.
What’s actually changing
The old rules governed export of a product. The new approach goes after a service.
- Before: Controls covered selling or shipping restricted chips to China.
- The gap: Remote access through third-country cloud infrastructure sidestepped those controls entirely.
- Now: A rule that treats renting compute to Chinese entities much like exporting the chip itself.
What stands out here is the recognition that in the AI era, access to compute matters more than ownership of it. A frontier model doesn’t care whether the GPUs sit in a rack you own or one you rent by the hour.
Why it matters for the industry
This reaches well beyond China policy. If the rule lands the way The Information describes, it touches every cloud provider that serves international customers.
- Cloud giants face new compliance weight. Providers may need to verify who’s really behind an account, where they’re based, and who ultimately benefits from the compute. Know-your-customer rules, but for GPUs.
- Overseas data centers come under scrutiny. Facilities in Southeast Asia and the Middle East that have absorbed huge chip orders could face questions about their end users.
- Nvidia and its rivals feel the pressure again. Remote-access demand has been one channel keeping Chinese buyers connected to restricted hardware. Cutting it narrows the market further.
The bigger pattern
This fits a multi-year tightening that began under the Biden administration and keeps escalating. First came limits on the fastest chips. Then Nvidia designed slower China-specific parts, and those got restricted too. Then came controls on chipmaking equipment. Each step closed one door, and each time the trade found another.
Remote access was the obvious next door. Washington closing it signals that officials now view AI compute as strategic infrastructure, on par with the chips themselves. The question they’re answering: what good is blocking the export if the capability flows through a browser tab?
What to watch next
The rule isn’t final, and the details will decide how much it bites. Keep an eye on a few things:
- How “remote access” gets defined. Broad language could sweep in ordinary cloud services. Narrow language could leave new loopholes.
- Enforcement burden. Verifying end users across global data centers is hard. Expect pushback from cloud providers on cost and feasibility.
- China’s response. Beijing has answered past controls with its own restrictions on rare earths and other inputs. Another escalation is likely.
- Timing. Draft rules can take months to finalize and often shift under industry lobbying.
For practitioners, the practical takeaway is simple: if your company runs AI workloads through international cloud infrastructure, compliance is about to get more complicated. Legal and procurement teams will want to track this closely.
The US has spent three years trying to keep its best AI hardware out of Chinese hands. This latest step admits a hard truth: in a cloud-first world, controlling the chip was only ever half the job. Full details are available at the original report from The Information.