US Closes ‘Cloud Loophole’ on AI Chips in New Rule Aimed at China
The United States has moved to close the gap that let Chinese companies rent America’s most powerful artificial intelligence chips without ever buying one. The Commerce Department’s Bureau of Industry and Security issued an interim final rule on Monday, 16 November, extending export controls from physical shipments to remote access through data centres abroad.
For four years, Washington has banned shipments of its most advanced AI chips to China. But a Chinese firm could still rent computing power from a data centre in a third country, running models on American hardware it was never allowed to own. The new rule is designed to end that arrangement.
“The rule is designed to ensure that existing restrictions on access to sensitive American technology cannot be circumvented through remote computing arrangements involving data centers in third countries,” BIS said in its announcement.
A threshold measured in computing power
At the heart of the rule is a single number: 12 million aggregate total processing performance, or TPP, the export-control measure of computing power. A licence will be required when an exporter knows that covered chips or servers will support a data centre — anywhere in the world — giving more than that level of computing access to a company headquartered in Macau or a Country Group D:5 destination, including China, or whose ultimate parent is based there.
That is roughly the maximum output of about 750 of Nvidia’s H100 or H200 chips. The rule also covers indirect access through intermediaries or brokers, and linked facilities that share resources as a single system. Operators will have to check their customers’ ownership structures to make sure they are not ultimately run from a restricted country.
“This new control is a logical extension of existing controls,” the rule states. “These restrictions will undermine the ability of companies headquartered in countries of concern from using U.S. advanced Super Intelligence (SI) semiconductors to train advanced SI models.”
Data centres already providing covered computing access to restricted firms will have 90 days from the effective date to obtain a licence. Applications under the new data-centre end-use control will be reviewed case by case.
Five countries face new licence requirements
Separately, the rule adds licence requirements for advanced semiconductor exports to or within five countries the Commerce Department says present a higher-than-normal risk of diversion: Indonesia, Malaysia, Singapore, Thailand and Brazil.
US-headquartered recipients, or those with a US-headquartered ultimate parent, are excepted, subject to reporting requirements. BIS said the rule was written by and at the direction of the Treasury Department and the National Security Council.
There is one narrow carve-out: computing access used exclusively for commercial recommender systems — the kind that rank products or posts for shoppers and social media users — is excluded, subject to reporting conditions. It does not cover systems designed to generate new text, images, audio, video or software.
The Nvidia scandal behind the timing
The rule lands days after Nvidia co-founder Jensen Huang resigned as chief executive amid revelations about his relationship with a woman alleged to be a Chinese spy linked to one of the chipmaker’s customers. Washington has reportedly been considering placing that customer, Megaspeed, on a special entity list, amid claims it bought thousands of Nvidia chips to supply computing power to Chinese businesses.
Treasury Secretary Scott Bessent has given Nvidia’s board 10 days to explain whether its most advanced chips reached China, and whether it used shell companies, many based in Singapore, to get around the rules. Reuters reported last week that Chinese AI companies were rushing to secure overseas computing capacity before tighter restrictions arrived.
The rule was published as President Donald Trump began an Asia trip ahead of a meeting with Chinese President Xi Jinping.
An old framework is formally buried
The interim rule also formally rescinds the Biden administration’s AI Diffusion Rule, issued in January 2025, which sorted the world into three tiers for chip access and restricted transfers of certain powerful AI models. BIS said the framework was never enforced, and the Trump administration announced in May that it would scrap it.
It replaces country tiers with targeted destination, end-use and remote-computing restrictions aimed at the routes officials say were used to evade controls.
What happens next
The rule takes effect on publication in the Federal Register, for which no date has been set. Comments will be due 60 days after publication, and the 90-day transition for existing operators runs from the effective date.
For data centre operators and chip exporters, the question is no longer only where the hardware goes, but who ultimately uses the computing power it provides. If a data centre rents large-scale AI capacity to a company whose parent sits in Beijing, Washington now wants a licence first.