Policy

US Commerce Extends AI Chip Export Licenses to Overseas Data Centers Serving China

Washington now treats rented computing power as an export, forcing cloud operators worldwide to vet who ultimately benefits from American silicon.

T
By TechQuire Daily Staff TechQuire Daily Staff
November 16, 2026 / 7 min read

For most of the past decade, American export controls on advanced artificial intelligence hardware were built around a simple physical question: where is the chip going? That question became harder to answer as cloud computing matured, because a chip can sit in a rack in Singapore or Kuala Lumpur and still be rented, hour by hour, by a company headquartered in Shenzhen. Washington has now decided that the answer matters as much as the shipping label, and on November 16, 2026 it moved to close the distance between the two.

The Bureau of Industry and Security, the export enforcement arm of the US Commerce Department, published an interim final rule that extends licensing requirements to remote access, not merely to physical shipments. The measure is filed as RIN 0694-AK50 and was, in the agency's own framing, written by and at the direction of the Department of the Treasury and the White House National Security Council. That provenance is unusual, and it signals that the rule is as much a political document as a technical one.

The gap it targets is real. US rules already bar the sale of the most advanced chips to China, Russia and other restricted destinations, but they did not clearly cover a Chinese firm renting equivalent computing power from a data center in a third country. Several Southeast Asian operators built businesses on exactly that arbitrage, and the arrangement allowed restricted buyers to train large models on American silicon without ever taking delivery of it.

The rule lands at a delicate moment. It is the first formal policy response to a scandal involving Nvidia and Megaspeed, a Singapore-based cloud provider, and it arrives days before President Donald Trump is due to travel to China for an economic summit with Xi Jinping. Markets noticed: shares of Nvidia, Micron Technology, Intel and Advanced Micro Devices all fell on Monday morning after the announcement.

Key Facts

Under the interim final rule, a license is required when a data center anywhere in the world provides more than 12 million aggregate total processing performance, or TPP, of advanced computing power to companies headquartered in Macau or in Country Group D:5. TPP is the export control metric Washington uses to size computing power. TechCrunch reported on November 16 that 12 million TPP is roughly equivalent to the peak computing power of about 750 Nvidia H100 or H200 chips, which puts the threshold within reach of a mid-sized commercial cluster.

Country Group D:5 is the embargoed list at the centre of the controls. CNBC reported on November 16 that the group includes China, Russia, Iran and North Korea alongside 18 others, for 22 countries in total. The rule treats remote access as a logical extension of the shipment controls that already applied to those places, closing what BIS described as a route around restrictions on sensitive American technology.

The rule also adds licensing requirements for exports, reexports and in-country transfers of advanced semiconductors to five destinations: Indonesia, Malaysia, Singapore, Thailand and Brazil. Commerce said those countries present a higher-than-normal risk of diversion. Shipments destined for US-based companies are carved out, according to TechCrunch.

Coverage is deliberately broad. It reaches indirect access arranged through brokers and go-betweens, and it obliges exporters and data center operators to examine customers' ownership structures and activities rather than taking a name on a contract at face value. Euronews reported on November 16 that overseas data centers will be required to conduct due diligence on their customers. Existing operators that provide covered services must obtain licenses within 90 days of the rule taking effect.

Two further details matter for industry. The rule does not apply to commercial recommendation systems such as those used in e-commerce or social media, a carve-out that keeps ordinary consumer platforms outside the perimeter. It also rescinds the Biden administration's AI Diffusion Rule from January 2025, which had sought country-level caps on advanced AI chip exports. The new approach is transaction and customer focused rather than quota based.

Analysis

The immediate question is why the rule was drafted at Treasury and the National Security Council rather than inside Commerce. CNBC reported on November 16 that the direction came from those two bodies, and that the move followed mounting criticism of Commerce Secretary Howard Lutnick for not doing enough to stop possible diversion of Nvidia's most advanced chips to China. Treasury Secretary Scott Bessent, in a Friday interview with CNBC's Squawk Box, blamed Commerce for failing to place Megaspeed on its entity list.

The Megaspeed thread runs through the whole episode. The New York Times reported on November 9 that US authorities began investigating Megaspeed and executive Alice Huang in 2024, after the company started importing billions of dollars' worth of advanced Nvidia AI equipment into Malaysia, and that Megaspeed had been building data centers around Southeast Asia to serve Chinese clients including Alibaba. Megaspeed has rejected allegations that it sent chips to China. The affair escalated into a corporate crisis: Nvidia chief executive Jensen Huang resigned last week after reports that he had an affair with an alleged Chinese spy who worked at Megaspeed, one of Nvidia's biggest Asian customers.

What this really means is that the enforcement problem has moved from customs paperwork to contract terms and customer files. A shipment control can be checked at a port. A remote access control cannot, which is why the rule leans on due diligence obligations for data center operators and on ownership scrutiny that reaches through layered corporate structures. That is a far harder regime to police, and it places a compliance burden on cloud providers in countries that are not themselves subject to US embargoes.

There is a strategic trade off embedded in the design. By rescinding the AI Diffusion Rule and replacing country caps with a threshold pegged to computing volume and customer nationality, the administration is signalling that it wants legitimate commercial adoption of American technology to continue abroad. BIS said explicitly that the United States remains committed to promoting global adoption of American technology for legitimate commercial purposes. The bet is that a narrower, better targeted control can be enforced more credibly than a broad country-based ceiling.

Why It Matters

For Nvidia and its peers, the rule adds a layer of diligence to sales that previously looked routine. The threshold is expressed in aggregate computing power rather than in a named list of customers, so a cluster assembled for regional enterprise clients can fall inside the perimeter depending on who signs the contracts. Buyers in Singapore, Malaysia and Thailand now face licensing questions that did not exist a week ago.

For Southeast Asia's data center boom, the effect could be material. Countries that positioned themselves as neutral hosts for regional AI workloads now sit in a category that Commerce says carries a higher-than-normal risk of diversion. Operators will need to document ownership chains, and the 90 day window for existing providers is short relative to the procurement cycles involved.

For the broader control regime, the precedent is the important part. If remote access counts as export, then the physical location of a server stops being the decisive fact in export control law. Chris McGuire, a former deputy senior director for technology and national security at the National Security Council, called the rule a major step forward to countering Chinese smuggling of and access to advanced AI chips, according to CNBC. That endorsement suggests the direction of travel will survive changes in personnel.

Next Up

The rule is an interim final rule, which means it takes effect while the comment process runs, and its 90 day compliance clock starts the countdown for data center operators that already provide covered services. How BIS defines covered services in practice, and how aggressively it pursues brokers and go-betweens, will determine whether the threshold holds.

Diplomacy will run in parallel. Trump's economic summit with Xi Jinping is days away, and the rule gives Washington a concrete enforcement action to point to. Whether Beijing reads the rescission of the AI Diffusion Rule as a concession, or the new licensing requirements as an escalation, will shape the next round of the dispute.

Tagged

Comments (0)

No comments yet. Be the first to share your thoughts.