The Wire
Washington closes the cloud loophole on China’s AI chips as Trump flies to meet Xi
A new Commerce Department rule forces overseas data centres to seek US licences before renting advanced computing power to Chinese firms; and names five countries as diversion risks, days after the Nvidia scandal shook the chip debate.
Key points
- On Monday 16 November 2026, the US Commerce Department’s Bureau of Industry and Security issued an interim final rule requiring licences for data centres that supply more than 12 million aggregate TPP of computing power to companies headquartered in China, Macau or other Country Group D:5 destinations. CNA; TechCrunch, 16 Nov 2026
- Exporters will also need US licences to ship advanced AI chips to Singapore, Thailand, Malaysia, Indonesia and Brazil, cited as presenting a higher-than-normal risk of diversion to China. Nikkei Asia; ANI, 16 Nov 2026
- The rule formally rescinds the Biden-era AI Diffusion Rule of January 2025, which was never enforced. Commerce Department; CNN, 16 Nov 2026
- The announcement came as President Donald Trump flew to Asia to meet President Xi Jinping in Shenzhen on Thursday, at the APEC summit. CNA, 16 Nov 2026
- Nvidia chief executive Jensen Huang resigned on Thursday after admitting to an affair with an executive at Megaspeed, a Singapore data-centre company under scrutiny for possible diversion of chips to China. CNA; CNN, 16 Nov 2026
ONE EVENT. THREE READINGS.
The story reads as a long-overdue repair of a leaking regime.
Chris McGuire, a senior fellow at the Council on Foreign Relations, called the rule “the first good-faith effort to actually close loopholes in export controls in a year and a half”.
What this reading highlights: The Western debate is about enforcement, not principle: whether the Commerce Department, which spent months resisting the rule, can now administer it; and whether the licensing queue becomes a real barrier or a rubber stamp.
The scandal supplies the political energy: the old argument for selling chips to China was compromised by the very companies making it.
Sources: CNN (1), TechCrunch (1)
The story lands differently: it is a rule written in Washington that names Asian countries as the problem.
Singapore, Thailand, Malaysia and Indonesia, four of the five countries named, are the region’s data-centre hubs, and the rule puts their operators under American licensing.
What this reading highlights: Nikkei Asia, reporting from Hong Kong, stressed Chinese AI companies’ fears that Huang’s downfall would accelerate American restrictions on their access to computing power.
For the region, the question is whether South-East Asia’s digital economy must now be governed from Washington; and what that costs the countries that built it.
Sources: CNA (1), Nikkei Asia (2)
The rule reads as a warning to every country that hosts the infrastructure of the technology war.
Brazil is named alongside the Asian hubs; India, which hosts a growing data-centre industry of its own, is watching closely; Indian outlets carried the rule in full, noting the due-diligence burden on data centres and the case-by-case review of licences.
What this reading highlights: The South’s stake is practical, not ideological: the servers are built on its soil, the jobs are in its cities, and the rules are now written in Washington.
The Blackwell study shows the South’s infrastructure is already a front line, whether it chose that role or not.
Sources: Asianet News (ANI) (1), Zee Business (1), The Indian Express (1)
Sources by lens: 2 West, 3 East, 3 Global South. See every source
THE POINT OF CONVERGENCE
On Monday 16 November 2026, the United States Commerce Department’s Bureau of Industry and Security issued an interim final rule requiring export licences for data centres that supply more than 12 million aggregate TPP of computing power to companies headquartered in China, Macau or other Country Group D:5 destinations; exporters will also need licences to ship advanced AI chips to Singapore, Thailand, Malaysia, Indonesia and Brazil; the rule formally rescinds the Biden-era AI Diffusion Rule; and it was announced as President Donald Trump flew to Asia to meet President Xi Jinping in Shenzhen on Thursday.
WHAT REMAINS CONTESTED
On whether the rule will work. Supporters, including the Council on Foreign Relations, call it the first serious attempt to close the loopholes in a year and a half. Critics in industry warn that licensing will burden allied data-centre economies and push Chinese demand towards smuggling and domestic chips. And the administration itself is split; the Treasury and the National Security Council wrote the rule over the Commerce Department’s resistance.
The story
What Washington announced on the runway to Asia
The Trump administration said on Monday, 16 November 2026, that it will further restrict China’s access to the best artificial-intelligence chips made by companies like Nvidia, announcing the move as President Donald Trump began a trip to China to meet President Xi Jinping.
The rule, issued by the Commerce Department’s Bureau of Industry and Security, closes the route Chinese companies have used most since direct exports were banned: renting computing power from data centres in other countries. Any data centre, anywhere in the world, that supplies more than 12 million aggregate TPP, total processing performance, the government’s measure of computing power, to a company headquartered in China, Macau or another country in the restricted Country Group D:5 will now need a United States export licence. Data centres already providing that much computing power have 90 days to obtain a licence or face further export restrictions.
The rule also names five countries, Singapore, Thailand, Malaysia, Indonesia and Brazil, where exporters will now need licences to ship advanced chips, because they present a higher-than-normal risk of the chips being diverted to China. And it formally rescinds the Biden-era AI Diffusion Rule of January 2025, which sorted the world into tiers for chip access but was never enforced.
The Commerce Department said the rule was written by the Department of the Treasury and the National Security Council, which also directed its release. Trump is due to meet Xi in Shenzhen on Thursday, at the APEC summit, where chips are expected to be discussed. (Thursday’s Trump–Xi–Putin summit)
How the remote-access route actually worked
Since the United States began restricting advanced chip exports to China in October 2022, Chinese technology companies have rented computing capacity from data centres in South-East Asia to train their models on servers outside mainland China. The chips stayed in Malaysia or Thailand; the computing power crossed the border by cable. That arrangement was legal; until Monday.
The new rule is designed, in the Commerce Department’s words, to ensure that existing restrictions “cannot be circumvented through remote computing arrangements involving data centres in third countries”. The licence requirement will not apply to systems used by online retailers and social-media companies, which the department said do not pose the same national-security threat. Applications will be reviewed case by case, and data centres will have to vet their customers, including examining direct and indirect ownership by entities based in restricted destinations.
The announcement follows weeks of scandal around Nvidia, the company at the centre of the chip trade. Its chief executive, Jensen Huang, resigned on Thursday after admitting to an affair with an executive at Megaspeed, a Singapore-based cloud company and Nvidia partner under scrutiny for possible illegal diversion of chips to China. The Commerce and Justice Departments began investigating Megaspeed in 2024. On Friday, Treasury Secretary Scott Bessent said four government agencies had favoured putting Megaspeed on a trade blacklist, but the Commerce Department, which was in charge, had not. Megaspeed said on 15 November that it complies with all applicable laws, including United States export-control requirements.
Days earlier, a study by Will DePue, a former OpenAI researcher, found that six open-weight AI models released since April by four Chinese companies, DeepSeek, Moonshot AI, MiniMax and Ant Group, carry numerical patterns suggesting they were trained on Nvidia’s Blackwell chips, which are barred from export to China. The finding, reported by The Indian Express, sharpened the debate over whether the old controls were working at all. (Nvidia as national champion)
cannot be circumvented through remote computing arrangements involving data centres in third countries.
What the rule leaves unanswered
Several facts that will decide whether the rule bites are still unknown. The rule takes effect on publication in the Federal Register; it had not appeared there by Monday afternoon, though one report said publication could come as early as 19 November. Comments are due 60 days after publication.
It is not yet clear how strictly licence applications will be judged, or how many will be granted. The department said applications would be reviewed case by case, but has not said what makes a data centre a trusted customer.
Nor is it known how Beijing will respond. China has been tightening its own controls on AI exports, and its foreign ministry warned last week against using the Nvidia affair as a pretext for new restrictions. Whether the rule survives Thursday’s meeting in Shenzhen is also open. (Why China chose Shenzhen for APEC)
Finally, the fate of Megaspeed itself is unresolved. The company has not been placed on the Entity List, and the administration is divided over whether it should be; a split Trump himself described on Monday as two acceptable views: “One wants to be friendly with China and one wants to have a blowup with China.”
Why a licence form in Singapore reaches into every AI lab in China
Computing power is the bottleneck of the artificial-intelligence race. Training a frontier model needs thousands of the most advanced chips running together for weeks, and the United States makes nearly all of them. Whoever controls access to that computing power controls the pace at which rival labs can build. By extending the controls from shipping chips to renting computing power, Washington has moved the technology war from the port to the cloud.
The five named countries are now on the front line. Singapore, Malaysia and Thailand built large data-centre industries by hosting the world’s computing capacity; the rule forces their operators to choose between Chinese customers and American chips. Brazil and Indonesia, which host growing data-centre capacity of their own, face the same choice. Governments in the region will have to decide how much of their digital economy they are willing to put under Washington’s licensing regime.
For China, the stakes are direct. Its leading AI companies have relied on rented foreign computing power to train models on hardware they cannot legally buy. If the licences are refused, their training runs slow down; or move to domestic chips still catching up.
The pendulum swings; from selling chips to sealing the cloud
The rule is, in substance, a reversal. For most of the past year, the administration’s public argument, led by Huang himself, was that selling chips to China was the strategy: American technology would set the standard, and the revenue would fund the next generation. The revelations of the past week put that argument into question, as Kate Koren, a former Commerce official, put it: “The pendulum is swinging in the other direction; towards tighter export controls.”
The rule also settles, for now, an open fight inside the administration. Bessent, Secretary of State Marco Rubio and Trade Representative Jamieson Greer had been pushing for the rule; the Commerce Department, under Howard Lutnick, had resisted. The department’s own statement gives the game away: the rule was written by the Treasury and the National Security Council, not by Commerce. Lutnick is travelling with Trump to Asia, and the president said on Monday that he has no plans to replace him. The hawks won the rule; Lutnick kept his job.
But a rule written by one department at the direction of another, announced on the eve of a summit, can be traded at that summit.
What happens next if the licences are refused
If the licences are refused in practice, three shifts follow. First, Chinese labs will accelerate work on domestic chips and on training methods that need less computing power; a push Beijing was already making. Second, data-centre investment in the five named countries will slow, or shift towards customers whose ownership raises no questions in Washington. Third, the smuggling routes will adapt: the history of export controls is a history of evasion, and the Commerce Department itself warned that countries tolerating evasion could face import restrictions or a downgrade in their export status. The rule does not end the contest over computing power; it moves it into the licence queue, and from there into the shadows.
What Thursday’s handshake in Shenzhen could change
Four things to watch. First, the Federal Register: publication starts the 90-day clock for existing data centres and the 60-day comment period. Second, Thursday’s meeting between Trump and Xi in Shenzhen, where chips are expected to be on the agenda; and where the rule could become a bargaining chip. Third, Beijing’s formal response, which will show whether China treats the rule as a negotiating position or as a lasting barrier. Fourth, the first licence decisions: they, more than the announcement, will tell the world whether the cloud loophole is closed or merely narrowed.
What we agree on, and what remains open
Established
- On Monday 16 November 2026, the United States Commerce Department’s Bureau of Industry and Security issued an interim final rule requiring export licences for data centres that supply more than 12 million aggregate TPP of computing power to companies headquartered in China, Macau or other Country Group D:5 destinations.
- Exporters will also need licences to ship advanced AI chips to Singapore, Thailand, Malaysia, Indonesia and Brazil.
- The rule formally rescinds the Biden-era AI Diffusion Rule.
- It was announced as President Donald Trump flew to Asia to meet President Xi Jinping in Shenzhen on Thursday.
Contested
- On whether the rule will work. Supporters, including the Council on Foreign Relations, call it the first serious attempt to close the loopholes in a year and a half. Critics in industry warn that licensing will burden allied data-centre economies and push Chinese demand towards smuggling and domestic chips. And the administration itself is split; the Treasury and the National Security Council wrote the rule over the Commerce Department’s resistance.
Still open
- When the rule will appear in the Federal Register.
- How strictly licence applications will be judged.
- How Beijing will respond.
- Whether Thursday’s meeting in Shenzhen will change the rule.
- Or whether Megaspeed will be placed on the Entity List.
Questions, answered
Since the United States banned direct exports of advanced AI chips to China in October 2022, Chinese companies have rented computing power from data centres in other countries instead. The chips stayed in places like Malaysia or Thailand, while the computing power crossed the border by cable. The new rule requires a United States licence for any data centre supplying more than 12 million aggregate TPP of computing power to companies based in China or other restricted countries.
TPP stands for total processing performance, the United States government’s measure of a chip’s computing power. The threshold is set in computing power, not in a number of chips, so it covers any combination of hardware that adds up to that level of performance. A data centre crosses the line when the computing power it supplies to companies in China, Macau or other restricted destinations exceeds 12 million aggregate TPP, whether at one site or across digitally linked facilities.
Singapore, Thailand, Malaysia, Indonesia and Brazil were identified as destinations presenting a higher-than-normal risk that advanced chips would be illegally diverted to China. Exporters will now need United States licences to ship advanced AI chips and servers to those countries. The rule does not ban shipments outright; licence applications will be reviewed case by case. Shipments to companies headquartered in the United States, or whose ultimate parent is, are exempt provided exporters meet specified reporting requirements.
No. The interim final rule takes effect when it is published in the Federal Register, which had not happened by Monday afternoon; one report said publication could come as early as 19 November. Once it takes effect, data centres already supplying that much computing power to restricted companies have 90 days to obtain a licence. The public then has 60 days to submit comments, and the rule could still be revised after that process.
MAGNA BUREAU