The Wire
Washington’s new chip licences put Southeast Asia’s AI ambitions on probation
The United States will require licences for advanced AI chips shipped to Singapore, Malaysia, Thailand, Indonesia and Brazil, and for data centres renting large-scale computing power to Chinese firms; a move that stalls projects, rattles markets and pushes the region to build the export controls it never had.
Key points
- The US Bureau of Industry and Security announced an interim final rule on 16 November 2026 requiring licences for advanced AI chip shipments to Singapore, Malaysia, Thailand, Indonesia and Brazil. BIS, via CNA, 17 Nov 2026
- Data centres anywhere in the world that provide more than 12,000,000 aggregate TPP of computing power to China-headquartered companies must obtain a US licence, with existing operators given 90 days. Katten, 17 Nov 2026
- The rule takes effect when it is published in the Federal Register, scheduled for 19 November 2026, and formally rescinds the Biden-era AI Diffusion Rule. Nextgov, 16 Nov 2026
- Thailand drew US$9 billion and Malaysia US$4 billion of data-centre investment in the first three quarters of 2025, among the largest inflows in the world. UN Trade and Development, via CNA, 17 Nov 2026
- Nvidia shares fell 4.1% on Monday 16 November and have shed more than 30% of their value over the past month; the Nasdaq fell 2.5%. Al Jazeera, 17 Nov 2026
ONE EVENT. THREE READINGS.
The rule reads as the closing of a loophole that should never have stayed open.
American reporting has treated the Megaspeed affair as the trigger: a Singapore company that bought more than US$4.8 billion of Nvidia chips in about two years, some of which American officials suspect were smuggled into China rather than installed in its data centres.
What this reading highlights: The Western frame is enforcement, export controls are only as strong as their weakest transit point, and the market reaction sharpened the point: the Nasdaq fell 2.5 percent on Monday, Nvidia 4.1 percent.
But the West is not of one mind about who wrote the rule: Al Jazeera notes it was drafted by the Treasury Department and the White House, not Commerce Secretary Howard Lutnick’s department, and a public rift has opened between Lutnick and Treasury Secretary Scott Bessent over how strictly to limit China’s access.
The story is about a region being asked to choose; and refusing to say so out loud.
CNA’s analysis put it plainly: for more than a decade, Southeast Asian countries have told Washington and Beijing the same thing, don’t make us choose sides.
What this reading highlights: The rule makes that position harder to hold, because every licence application becomes a small declaration of alignment.
The regional press has focused on the practical damage: projects delayed, financing paused while lenders wait to learn when chips will arrive, and a reputation as a trusted AI hub put at risk just as the investments were arriving. Yet the East also sees an opening: the rule could push governments to build the export-control regimes they have postponed for years.
Sources: CNA (1), The Business Times (1)
The rule lands as a decision taken elsewhere about infrastructure the South is building.
Al Jazeera, reporting from Doha, and Free Malaysia Today, reporting from Kuala Lumpur, both stressed the same asymmetry: the five named countries did not write this rule, but their data-centre industries must now live under it.
What this reading highlights: Malaysia and Indonesia, where the data-centre boom has been fastest, carry the heaviest exposure; and the least say.
Brazil’s inclusion underlines that the map is not only Asian. What the South watches now is whether the rule becomes a template; a standing American veto over who may compute where.
Sources: Al Jazeera (1), Free Malaysia Today (1)
Sources by lens: 2 West, 2 East, 2 Global South. See every source
THE POINT OF CONVERGENCE
On the core facts; the Bureau of Industry and Security announced the rule on 16 November; five countries are named; licences are now required both for shipments of advanced chips and for large-scale remote access by China-headquartered firms; existing data centres have 90 days; the rule takes effect on Federal Register publication, scheduled for 19 November; and American-headquartered firms are exempt, provided they file monthly reports.
WHAT REMAINS CONTESTED
On whether the controls will work. Washington presents them as the closing of a loophole. Regional operators and analysts warn that licences will be slow, projects will stall and financing will wait. Others argue the real effect is political; a bargaining chip handed to Washington days before President Trump meets President Xi.
The story
What Washington ordered on Monday
Singapore; The United States moved on Monday to close the last open door through which Chinese companies have been reaching its most advanced artificial intelligence chips: the data centres of Southeast Asia. The Commerce Department’s Bureau of Industry and Security announced an interim final rule, ‘Addressing Remote Access to Advanced Semiconductors’, requiring American licences for shipments of advanced AI chips to Singapore, Malaysia, Thailand, Indonesia and Brazil, and for any data centre, anywhere in the world, that rents large-scale computing power to companies headquartered in China.
The rule, filed for publication in the Federal Register on 19 November, takes effect on publication. Washington’s stated reason is diversion: the bureau said the five countries present an ‘unacceptable risk’ of controlled chips being illegally re-routed to China.
The timing is pointed. President Donald Trump was leaving Washington for Asia as the rule was announced; first Manila, for talks with Southeast Asian leaders, then China, where he meets President Xi Jinping on Thursday. The announcement also lands in the middle of the scandal that has consumed Nvidia: its chief executive, Jensen Huang, resigned last week after reports of an affair with a former executive of Megaspeed, a Singapore-based data-centre company that American officials suspect of funnelling Nvidia chips to China. Beijing has called the spying allegations ‘entirely groundless’. The affair has already produced a criminal case, and the scandal has been driving the chip debate for days.
What the rule actually requires; and who it hits hardest
The rule has two working parts. First, the destination rule: any shipment of covered advanced-computing items, generally advanced AI chips and servers, to Indonesia, Malaysia, Singapore, Thailand or Brazil now requires a US export licence. American-headquartered recipients are exempt, provided exporters file monthly reports with the bureau. Second, the remote-access rule: a licence is required whenever a supplier knows a data centre will provide more than 12,000,000 aggregate total processing performance, the combined computing power of the chips, measured across linked sites, to or for a company headquartered in China, Macau or another restricted destination. Data centres already serving such customers have 90 days from the rule’s effective date to obtain a licence, a window that runs to mid-February.
The companies most exposed are the region’s AI cloud operators and data-centre owners that are not American-owned and that serve Chinese customers. ‘In Singapore, I’d add the traders and system integrators who buy chips and servers here and ship them on,’ Matthias Hendrichs, a Singapore-based adviser to AI companies, told CNA, ‘and anyone who wants to own their own chips, like telcos, research institutes and sovereign AI projects.’ Firms that rent capacity from an American cloud provider are far less affected, because the licence burden sits with the provider.
The exposure is uneven. Singapore and Malaysia have export-control laws or strategic trade management frameworks; Thailand has a framework it has not fully implemented; Indonesia has no dedicated regime and relies on scattered customs and sector-specific laws, according to Dr Maria Monica Wihardja of the ISEAS-Yusof Ishak Institute. That gap, in Washington’s reading, is why the region was flagged. Thailand and Malaysia drew US$9 billion and US$4 billion of data-centre investment in the first three quarters of 2025. The rule does not ban American chips from the region. It makes every shipment a permission slip, and every large Chinese customer a licensing event.
Beijing has called the spying allegations ‘entirely groundless’.
What is still unknown about the licences?
Almost everything that will decide whether the rule is a speed bump or a wall. Nobody outside the bureau knows how quickly licences will be granted, on what criteria, or how many applications will be refused. The bureau has said it will review applications case by case; but it has published no timetable and no approval rate.
Singapore has begun talks with Washington about trusted access for companies based there, its ministries said on Tuesday, but there is no public indication of what such access would cover, or when it might be agreed. Nor is it clear how many projects will be cancelled rather than delayed: Hendrichs told CNA he expects ‘delays of months’ for new projects.
The criminal case is equally open. The Commerce Department and the Justice Department confirmed late on Monday that investigators have opened a criminal investigation into Megaspeed; the Justice Department has taken the lead role. What the investigation will establish, and whether charges follow, is unknown. Megaspeed has said it conducted ‘legitimate and lawful business’.
Why does a licence rule in Washington reach into Asian server rooms?
Because the map of artificial intelligence is now drawn in server rooms, and Southeast Asia spent five years becoming one of its busiest addresses. The region offered what both superpowers wanted: cheap power, open investment rules and governments determined to stay neutral. That neutrality was the business model; until this rule ended it for the most advanced hardware.
A licence requirement does not stop the flow of chips, but it inserts Washington into every large transaction; and it forces every operator to declare, in writing, who its customers are and where their computing goes. For Singapore, the issue is less the volume of trade than standing. ‘There’s no doubt that this is a speed bump for Singapore’s AI ambitions,’ Josh Gilbert, eToro’s lead analyst for Asia Pacific, told CNA. Brazil is the fifth named country, a reminder that the diversion map is global. And the timing, days before Trump meets Xi, turns a technical export rule into a bargaining chip in the largest trade negotiation in the world.
The licence is the message, not the paperwork
The technical details, the 12-million TPP threshold, the 90-day window, the monthly reporting for American firms, matter less than what the rule does to the structure of the market. Until Monday, a Chinese company barred from buying Nvidia’s most advanced chips could still legally rent the computing power of those chips from a data centre in Singapore or Johor Bahru. The chips stayed in Southeast Asia; the intelligence they produced went to China. Washington has now decided that renting is buying, and that the location of the server is no defence.
The rule’s second message is aimed at the region’s governments. By naming the absence of export-control regimes as the reason for the flag, Washington has made domestic legislation a condition of market access. The third message is the timing. Announced as Trump left for Asia, the rule arrives at the negotiating table already in force. Whether the controls work depends on a question no licence can answer: can the United States track where its chips actually compute, once they have been installed, resold, rented and re-rented across borders? The market’s answer, so far, is doubt: Asian indexes fell across the board on Tuesday, and the sell-off in chip stocks suggests investors expect more restrictions, not fewer.
What happens when the licences start arriving; or don’t
The first-order effect is delay: projects waiting on chips, lenders waiting on dates, operators waiting on licences. The second-order effects will decide the region’s AI map. If licences come slowly, investment will drift to places outside the rule’s reach, Japan, South Korea, Australia, the Gulf, or back to the United States, where no licence is needed at all.
Governments that spent a decade avoiding export-control legislation now have a deadline written by someone else, and the fastest way to shorten the licence queue is to build the regimes Washington wants to see. A third shift concerns China. Cut off from rented computing, Chinese AI developers will lean harder on domestic chips and on computing hosted in countries Washington has not named. Washington has already signalled that further designations of AI champions and controls are on the table.
Three dates that will show whether the rule bites
Three dates will show how hard the rule bites. The first is 19 November, when the rule is scheduled for publication in the Federal Register and takes effect; the 90-day transition for existing data centres starts running from that day. The second is Thursday, when Trump meets Xi in Shenzhen: watch whether the chips rule is traded, softened or hardened in the room. The third is mid-February, when the 90 days expire and data centres serving Chinese customers must hold licences or stop the service. Between now and then, watch Singapore’s trusted-access talks and the first licence decisions. If the licences flow and the laws follow, the region will have paid for its boom with paperwork. If they do not, the boom moves elsewhere.
What we agree on, and what remains open
Established
- On the core facts; the Bureau of Industry and Security announced the rule on 16 November.
- Five countries are named.
- Licences are now required both for shipments of advanced chips and for large-scale remote access by China-headquartered firms.
- Existing data centres have 90 days.
- The rule takes effect on Federal Register publication, scheduled for 19 November.
- American-headquartered firms are exempt, provided they file monthly reports.
Contested
- On whether the controls will work. Washington presents them as the closing of a loophole. Regional operators and analysts warn that licences will be slow, projects will stall and financing will wait. Others argue the real effect is political; a bargaining chip handed to Washington days before President Trump meets President Xi.
Still open
- How strictly the bureau will grant licences, or how many applications it will refuse.
- Whether Singapore’s talks on trusted access will produce an exemption, and what it would cover.
- How many data-centre deals will be cancelled rather than delayed.
- What the criminal investigation into Megaspeed will establish.
Questions, answered
The Commerce Department’s Bureau of Industry and Security announced an interim final rule requiring US licences for advanced AI chip shipments to Singapore, Malaysia, Thailand, Indonesia and Brazil. It also requires licences for any data centre, anywhere, that rents large-scale computing power to China-headquartered firms. The rule takes effect on publication in the Federal Register, scheduled for 19 November.
Five destinations are named: Singapore, Malaysia, Thailand, Indonesia and Brazil. Washington says they present an unacceptable risk that advanced chips will be illegally diverted to China. Shipments to those countries now need a US export licence, except where the buyer is headquartered in the United States and files monthly reports. The remote-access rule, by contrast, applies to data centres worldwide.
Megaspeed is a Singapore-based data-centre company that American officials suspect of funnelling Nvidia chips to China. It bought more than US$4.8 billion of Nvidia chips in about two years. Nvidia chief executive Jensen Huang resigned last week after reports of an affair with Alice Huang, a former Megaspeed executive alleged to have been recruited by Chinese intelligence. Commerce and Justice have opened a criminal investigation.
Operators that are not American-owned and serve Chinese customers are hit hardest: they will need licences both to buy new advanced chips and to keep serving those customers at scale, with 90 days to comply. Experts expect delays of months for new projects and financing to wait while lenders seek certainty. Firms renting capacity from American cloud providers are far less affected, since the licence burden sits with the provider.
MAGNA BUREAU