Skip to main content
← News ENFR SearchSubscribe

No single capital owns the whole truth

The Wire

Chinese tech firms fear AI hits from US curbs after Nvidia fallout

Washington’s new rule closes the remote-access route Chinese firms used to rent advanced computing abroad; and Alibaba and Tencent, reporting bumper AI quarters, say the squeeze is already real.

A copper smelter complex
A copper smelter complex; one link in the metals supply chain behind the world’s most advanced chips.Chris857, CC BY-SA 4.0, via Wikimedia Commons

Key points

  • The rule: On 16 November the US Bureau of Industry and Security issued an interim final rule requiring licences for advanced AI chips sent to Singapore, Malaysia, Indonesia, Thailand and Brazil, and for any data centre worldwide supplying more than 12 million units of computing performance to companies headquartered in China, Macau or other arms-embargoed states. It was published in the Federal Register on 19 November and took effect on publication. 2
  • Nvidia fallout: Jensen Huang resigned as Nvidia’s president and chief executive on 12 November, effective immediately, after admitting an affair with Alice Huang, a former executive at Singapore data-centre firm Megaspeed whom US intelligence believes to be a Chinese intelligence asset. Chief financial officer Colette Kress is interim chief executive. 3
  • Alibaba: AI Cloud and Compute Services revenue rose 53 per cent year on year to nearly 61 billion yuan (about $9 billion) in the July–September quarter; group revenue rose 10 per cent to almost 272 billion yuan. The company warned the new US rules could affect some of its global data-centre expansion plans and said it was still assessing the repercussions. 1
  • Tencent: President Martin Lau told analysts on 18 November that the new US rule on chip access ‘sets everyone back in China’, and that training timelines for the next versions of Tencent’s AI models would probably slip. Third-quarter revenue rose 8 per cent to 207.3 billion yuan. 4
  • Scale of the loophole: A study published in October by the Institute for Progress, a US think tank, estimated that remote access to AI chips through overseas data centres could account for roughly one-third of China’s AI compute by the end of 2027. 1

ONE EVENT. THREE READINGS.

West

In the Western reporting, the story is enforcement catching up with evasion.

The law firm Sidley reads the rule as a lawyer’s document, thresholds, grace periods, carve-outs, and notes it reaches chips already installed, not just new sales.

What this reading highlights: The Register leads with Tencent’s blunt verdict, and the Council on Foreign Relations treats Jensen Huang’s exit as a foreign-policy event, not a corporate one.

The Western lens asks whether the controls can be enforced without breaking the cloud business model they now regulate; and whether a licensing regime can keep pace with a technology that moves faster than the paperwork.

Sources: Council on Foreign Relations (1), The Register (1), Sidley (1)

East

In the Eastern reporting, the story is a map of the new front line.

CNA explains which Southeast Asian countries are named and why, and notes the rule was written by the Treasury and the National Security Council rather than the Commerce Department.

What this reading highlights: The Chosun Biz traces the Megaspeed revelations that forced Washington’s hand, and Zaobao reports Martin Lau’s warning to investors in Singapore and Hong Kong.

The Eastern lens stresses that the named countries are the region’s data-centre hubs; and that the rule’s real test will be whether licences flow or stall.

Sources: CNA (1), CNA Explains (1), Chosun Biz (1), Lianhe Zaobao (1)

Global South

In the Southern press, the story is a question of who decides.

Al Jazeera leads with the five named countries and Beijing’s charge of ‘economic coercion’, and notes that India, a major cloud market, is not among them.

What this reading highlights: The Global South lens asks what happens to the investment plans of the countries caught in between, and whether a licensing regime run from Washington leaves them any say over the computing power hosted on their own soil.

For the South, the rule is less about chips than about sovereignty over infrastructure.

Sources: Al Jazeera (1), The Week (AP) (1)

Sources by lens: 3 West, 4 East, 2 Global South. See every source

THE POINT OF CONVERGENCE

On the facts. On 16 November the US Bureau of Industry and Security issued an interim final rule licensing advanced AI chips to five countries and remote computing access for Chinese firms; Jensen Huang resigned as Nvidia chief executive on 12 November; Alibaba’s AI cloud revenue rose 53 per cent in the September quarter; and Tencent’s president said the rule ‘sets everyone back in China’. Every independent outlet reporting the story gives the same account.

WHAT REMAINS CONTESTED

On whether the curbs slow China down or speed it up. Washington frames the rule as closing a loophole; Beijing calls it ‘economic coercion’. Tencent’s Martin Lau says both things at once, training will slip in the short term, domestic chipmaking will accelerate in the long term, and analysts split on which effect dominates.

The story

Washington closes the rented-compute loophole

On Monday, 16 November, the US Commerce Department’s Bureau of Industry and Security announced an interim final rule that shuts the last open door in America’s chip blockade against China: renting the computing power instead of buying the chips. The rule was published in the Federal Register on Thursday, 19 November, and took effect on publication.

It does two things at once. First, it requires export licences for advanced AI chips sent to five countries Washington now designates as presenting an ‘unacceptable risk of diversion’; Singapore, Malaysia, Indonesia, Thailand and Brazil. Second, it requires a licence for any data centre, anywhere in the world, that supplies more than 12 million units of aggregate total processing performance, roughly the output of 300 of the most advanced commercial graphics processors, to companies headquartered in China, Macau or other countries under a US arms embargo. Data centres already serving those customers have 90 days to obtain a licence. The rule also formally rescinds the Biden-era AI Diffusion Framework of January 2025.

The timing is no accident. The rule landed four days after Jensen Huang resigned as chief executive of Nvidia, whose chips sit at the centre of the AI race, following reports of his relationship with a woman US intelligence believes to be a Chinese intelligence asset.

Advertisement

What the rule bans; and what it leaves open

Until now, the arrangement was legal. Chinese firms barred from importing the most advanced graphics processors could simply rent computing power from data centres in Singapore, Malaysia or Thailand, where the chips stayed put and the models were trained remotely. The new rule ends that: a licence is now required when the exporter knows the chips will feed a data centre supplying above-threshold compute to restricted customers; and the control reaches chips already installed, not just new shipments.

There are carve-outs. Companies headquartered in the United States are exempt from the new destination licensing, provided they file monthly reports. Computing supplied exclusively for commercial recommender systems, that ranks products and content, is also exempt, though systems that generate text, images or video are not.

The rule was written in an unusual place: by the Treasury Department and the National Security Council, rather than by Commerce, which normally runs export controls. And it arrives with legislation behind it; the Remote Access Security Act, which would make remote access a controlled activity in its own right, passed the House of Representatives 369 to 22 on 12 January and awaits the Senate.

First, it requires export licences for advanced AI chips sent to five countries Washington now designates as presenting an ‘unacceptable risk of diversion’.

How much compute does China actually lose?

No public figure says how much computing power flows to Chinese firms through overseas data centres today. A study published in October by the Institute for Progress, a US think tank, estimated that remote access could account for roughly one-third of China’s AI compute by the end of 2027; but that is a projection of a growing practice, not a measurement of the present, and it counts capacity that may never be built.

Nor is it clear how strictly Washington will hand out licences. The rule gives the Bureau of Industry and Security wide discretion, and the first approvals, or refusals, will set the real terms of the blockade. Beijing has denounced the measures as ‘economic coercion’ and promised ‘necessary measures’ in response, without saying what those would be.

Two other questions hang over the story: whether the Senate takes up the Remote Access Security Act, and what Nvidia does next. The company says it is no longer shipping chips including the H200 to China, even though Washington had earlier approved some sales; and with a permanent chief executive still to be chosen, its stance could yet shift.

The week the chip war reached the cloud

For four years the contest over AI chips was fought at the border: which processors could be sold, to whom, and under what licence. This rule moves the fight into the cloud, where the chips never cross a border at all. That matters because computing power, not data or talent, is now the binding constraint on building frontier AI models; and the country that controls the compute sets the pace of the race.

The timing sharpens the stakes. The rule was issued days before President Donald Trump is due to meet President Xi Jinping on the sidelines of the Asia-Pacific Economic Cooperation summit in Shenzhen, putting export controls at the centre of the two leaders’ conversation. It also lands in earnings season, when China’s two largest technology companies are reporting the strongest AI numbers of their histories; and warning, in the same breath, that the ground is shifting beneath them.

For the rest of the world, the rule redraws the map of the data-centre boom. Five countries that built their digital economies on hosting other people’s computing now need Washington’s permission to keep doing it.

Two earnings calls tell the same story from opposite ends

Alibaba reported first, on Thursday, 19 November. Its AI Cloud and Compute Services revenue rose 53 per cent year on year to nearly 61 billion yuan, about $9 billion, in the July-to-September quarter, while group revenue rose 10 per cent to almost 272 billion yuan. AI-related product revenue grew by triple digits for the thirteenth consecutive quarter, to 17.9 billion yuan. And yet the same company warned that the new American rules could affect some of its global data-centre expansion plans, and said it was still assessing the repercussions. Capital expenditure jumped 60 per cent to 50.5 billion yuan, as Alibaba pours money into infrastructure whose supply lines Washington now controls. Its chief executive, Eddie Wu, had called its new Zhenwu V900 chip the ‘most powerful’ AI chip in China in September, and the company launched its flagship Qwen4-Max model earlier this week.

Tencent was blunter. Its president, Martin Lau, told analysts on Wednesday that the rule ‘sets everyone back in China’, and that training timelines for the company’s next versions of its AI models would probably slip. Everyone, he said, is now working with less advanced compute, ‘no matter when you actually started leasing or building capacities outside of China’. But Lau also drew the line the industry now draws: inference, running finished models, should be covered by Chinese-made chips, and over the longer run every Chinese player will rely on domestic supply chains. Tencent’s own numbers underline the pressure: third-quarter revenue rose 8 per cent to 207.3 billion yuan, while capital expenditure surged 225 per cent to 65.1 billion yuan.

The paradox is now the consensus in Beijing’s tech circles: the curbs bite hardest in the short term, and accelerate self-sufficiency in the long term. The open question is how long the short term lasts; and whether China’s domestic chips can close the gap before the delays compound.

What happens to the countries in the middle?

The five named countries are not bystanders; they are the region’s data-centre hubs, and their investment plans now depend on American licences. Singapore, Malaysia and Thailand built their digital economies in part on hosting computing for Chinese customers. Brazil, the only non-Asian country named, faces the same licensing wall. For all five, the rule converts a commercial advantage, cheap power, cool climates, open markets, into a diplomatic negotiation.

The compliance burden is real. Exporters and data-centre operators must now monitor their customers’ ownership and activities, and the rule covers indirect access through brokers and intermediaries; the arrangements at the heart of the Megaspeed affair. American cloud providers, exempt as US-headquartered firms, stand to gain share in all five markets.

And the Megaspeed case itself is far from over. The US Justice Department has taken over a Commerce Department probe into whether the Singapore-based company helped Chinese firms sidestep restrictions, and Singapore opened its own inquiry in October. How those investigations end will set the template for enforcement, and the deterrent, worldwide.

Licences, lawsuits and a summit in Shenzhen

Three clocks are now running. Data centres already serving restricted customers have 90 days to secure licences. The public has 60 days to comment on the interim final rule, with comments due around 18 January 2027. And the political clock runs fastest of all: the Trump–Xi meeting in Shenzhen will test whether the rule is a bargaining chip or a permanent wall.

Watch, too, for the first licence decisions from the Bureau of Industry and Security, for Beijing’s promised ‘necessary measures’, and for Nvidia’s choice of a permanent chief executive. Interim chief Colette Kress inherits a firm whose most valuable customer base is now fenced off by its own government’s rules; and a board inquiry that is still open.

What we agree on, and what remains open

Established

  • On the facts. On 16 November the US Bureau of Industry and Security issued an interim final rule licensing advanced AI chips to five countries and remote computing access for Chinese firms.
  • Jensen Huang resigned as Nvidia chief executive on 12 November.
  • Alibaba’s AI cloud revenue rose 53 per cent in the September quarter.
  • Tencent’s president said the rule ‘sets everyone back in China’. Every independent outlet reporting the story gives the same account.

Contested

  • On whether the curbs slow China down or speed it up. Washington frames the rule as closing a loophole.
  • Beijing calls it ‘economic coercion’. Tencent’s Martin Lau says both things at once, training will slip in the short term, domestic chipmaking will accelerate in the long term, and analysts split on which effect dominates.

Still open

  • How strictly licences will be granted, how much compute actually flows through the overseas route, what Beijing’s promised ‘necessary measures’ will be, whether the Senate passes the Remote Access Security Act, or who will lead Nvidia permanently.

Questions, answered

Washington did not ban chip sales outright. Its Bureau of Industry and Security issued an interim final rule requiring licences for advanced AI chips sent to Singapore, Malaysia, Indonesia, Thailand and Brazil, and for any data centre worldwide supplying more than 12 million units of computing performance to companies headquartered in China, Macau or other arms-embargoed countries. Existing operators have 90 days to seek a licence.

Sources by lens

West (3)
4The Register(19 Nov 2026)
5Sidley(19 Nov 2026)
East (4)
1CNA(20 Nov 2026)
6CNA Explains(17 Nov 2026)
7Chosun Biz(17 Nov 2026)
8Lianhe Zaobao(19 Nov 2026)
Global South (2)
2Al Jazeera(17 Nov 2026)
9The Week (AP)(19 Nov 2026)

Open access ends December 31 — from January 1, 2027, the depth is for members.

See membership