US Closes Remote Computing Loophole to Limit China’s Access to AI Chips

Washington closed the remote-computing loophole this week, and China’s AI sector is bracing for impact.

The Commerce Department’s new rules restrict foreign data-center access to US chips, prompting warnings from Alibaba (NYSE: BABA), Tencent (HKG: 0700) and a shake-up at Nvidia (NASDAQ: NVDA) itself.

The timing matters. Alibaba had just reported another quarter of strong revenue growth for its international cloud business and AI compute infrastructure when the rules landed. Yet the company flagged that the new measures could affect some of its global data-center expansion projects, Noovo Info reported. It is still assessing the consequences.

The Bottom Line

  • The new rule applies immediately and requires a license for any data center worldwide serving clients headquartered in China, Macao, Russia, Iran or roughly twenty other embargoed countries, per Korben’s reporting on the interim rule.
  • The threshold is 12 million TPP of cumulative chip power — roughly 750 units of Nvidia’s flagship H100 — with operators given 90 days to obtain licenses and verify who sits behind each contract.
  • China’s Champions dilemma: Tencent’s president Martin Lau argues restrictions will accelerate domestic chip self-sufficiency, while roughly 600,000 Nvidia H200 GPUs sit blocked in Hong Kong warehouses, per Foro3D and Radio France.

The Loophole Washington Just Closed

Until now, US export controls stopped at physical delivery. Nothing prevented a Chinese company from renting identical compute in the cloud from a data center in Singapore or Malaysia. That window is shut.

Nvidia gpus atrapadas en hong kong sin salida
Photo: Foro3D

The Commerce Department published an interim rule applying immediately: a license is now required for any data center, anywhere, whose clients are headquartered in embargoed jurisdictions. Korben reported that the measure targets the gray market directly — Washington suspects Singaporean server-rental firm Megaspeed of buying thousands of Nvidia chips to serve Chinese clients, and Singapore police are investigating, with Malaysia examining the subsidiary Speedmatrix.

US-headquartered companies are exempt, on condition of monthly reporting. But the details of enforcement are the hard part, as Korben put it: regulators must now track thousands of rental contracts scattered across the world.

U.S. closes loophole allowing remote access to AI chips — CTV News

Huang’s Exit Reshapes the Lobbying Math

Jensen Huang, l’ancien patron de Nvidia, était l’un des plus ardents défenseurs des ventes de puces américaines à la Chine, et son départ, selon Radio France, change le rapport de force aux États-Unis, où républicains et démocrates envisagent de limiter encore plus les exportations de semi-conducteurs les plus performants, according to Noovo.

The measures themselves came via the Commerce Department, but Le Monde noted a press statement saying the document was “written by and at the request of” the Treasury Department and the National Security Council — an internal split made public after Treasury Secretary Scott Bessent publicly criticized Commerce Secretary Howard Lutnick over earlier easing.

New export authorizations now cover Indonesia, Malaysia, Thailand, Brazil and Singapore, countries deemed to carry above-normal diversion risk.

US Closes Remote Computing Loophole to Limit China's Access to AI Chips
Photo: Radio France

The consequences are already visible in warehouses and procurement records. Foro3D reported that roughly 600,000 Nvidia H200 GPUs remain stuck in Hong Kong awaiting Chinese import approval — even though Washington has authorized the export — because Beijing’s own authorities are slowing H200 imports to push domestic production. ByteDance and Tencent have received only small lots of about 10,000 units each, and Nvidia faces potential inventory write-downs of up to $6.6 billion.

In the gray market, prices have tripled, with an eight-GPU H200 server reaching around $900,000, Foro3D reported.

Metric Figure Source
License threshold (cumulative chip power) 12 million TPP ≈ 750 H100s Korben
Compliance window for existing operators 90 days Korben
H200 GPUs blocked in Hong Kong ~600,000 Foro3D / Radio France
Units received by ByteDance and Tencent ~10,000 each Foro3D
Nvidia inventory write-down risk up to $6.6 billion Foro3D

Enforcement gaps persist elsewhere. MEZHA reported, citing CNN’s review of procurement documents, that at least six blacklisted Chinese universities — including Beihang University, on the US entity list since 2001 — disclosed purchases of restricted Nvidia chips in recent years, two of them within recent months. “Rationally, you have to know that despite all your efforts, some will slip through,” said Kate Koren, per CNN’s reporting carried by MEZHA.

US Closes Remote Computing Loophole to Limit China's Access to AI Chips
Photo: Le Figaro

Alibaba’s quandary: growth at home, blockage abroad

Alibaba is not standing still. It launched its flagship Qwen4-Max AI model earlier this week, and in September the CEO of Alibaba called the new Zhenwu V900 chip the most powerful AI chip in China while announcing massive data-center investment plans at home and abroad.

But the revenue engine has a geography problem: Jacob Cooke said that Alibaba Cloud’s AI revenue comes largely from China, with significant untapped demand from Chinese enterprises remaining.

Tencent struck a sharper tone. Its Q3 results, released the same day, showed slowing revenue growth even as it pours money into AI — and its founder, Pony Ma, compared earlier efforts in AI to a ship “full of leaks,” adding, as Le Figaro reported, “We still hope the ship can sail faster.”

US Closes Loophole That Allowed China To Remotely Access Nvidia Chips — Forbes Breaking News

The Domestic Chip Bet — and Its Limits

China’s long game is substitution, and its executives say so plainly. “At long last, we believe all Chinese players will have to rely on domestic supply chains and chips. And we do see the sector progressing very rapidly in that area,” Martin Lau told analysts on Tencent’s earnings call.

But the Physics of substitution is unforgiving. Radio France reported that Huawei’s chips are on average about five times less powerful than Nvidia’s, though Huawei announced in September that its next-generation AI processors would be ready in the first quarter of 2027 — nine months earlier than expected — with Beijing’s backing.

Meanwhile, TSMC has tightened restrictions on Chinese chip designers, suspending some previously permitted advanced fabrication orders, per a report carried by Zonebourse: advanced Chinese designs are now presumed subject to US export controls, closing a manufacturing route that some Chinese AI chip designers had used to access 7nm production.

“At long last, we believe all Chinese players will need to rely on domestic supply chains and chips. We are seeing the sector progress very rapidly in this area.”
— Martin Lau, president of Tencent, on the group’s earnings call Wednesday

The CUDA software ecosystem adds another lock-in layer, Foro3D reported, keeping Chinese firms tethered to Nvidia’s architecture even as they wait. That dependency is why the blocked Hong Kong stockpile matters: the hardware exists, but neither customs nor software alternatives move easily.

Nvidia, the American technology company behind the H100 and H200 chips now piling up in Hong Kong warehouses, was founded in 1993 by Jensen Huang alongside engineers Chris Malachowsky and Curtis Priem, who first sketched out the venture in a meeting at a Denny’s diner in San Jose with $40,000 in the bank.

Originally a maker of graphics chips for video games — it released the GeForce 256, the first product marketed as a GPU, in 1999 — the company rode the AI boom to a historic milestone: in 2025 it became the first company to surpass US$4 trillion and US$5 trillion in market capitalization.

The next date to watch: 90 days from this week’s rule, when existing data-center operators must hold licenses or cut off restricted clients. By then, Alibaba says it expects to have finished assessing the impact on its global data-center expansion — and Nvidia’s most expensive warehouse will have collected another quarter’s dust.

Photo of author

Economy Desk

Economy Desk at Archyde.

HRC Names St. Petersburg, Mayor Welch National Standouts in Equality Index

Cowboys Demote Christian Parker From Playcalling Duties But Insist He Remains Long-Term Coordinator