Spectrum & PolicyExport ControlsAI ChipsData CentresBis

US closes AI chip loophole with 90-day licence ultimatum for data centres

BIS gave overseas data centres serving Chinese firms 90 days to obtain export licences, imposing a 12 million TPP cluster threshold and new licensing for five Asian and Latin American markets.

3 min read

US tightens AI chip rules, targets overseas data centresSpectrum & Policy
US tightens AI chip rules, targets overseas data centresAI-generated

Why it matters

  • BIS issued the interim final rule on 16 November, giving affected data centres 90 days to obtain a licence.
  • The rule applies to compute clusters above 12 million aggregate total processing performance, on one site or across linked facilities.
  • Advanced chips shipped to Indonesia, Malaysia, Singapore, Thailand and Brazil now require licences.
  • Jensen Huang resigned as Nvidia CEO on 12 November, with CFO Colette Kress named interim chief.
  • The IFR scrapped the Biden-era AI Diffusion Rule, including its country tiers and controls on AI model weights.

The story

Data centres anywhere in the world now have 90 days to obtain a US export licence or face supply blocks, under new rules targeting Chinese companies that rent American computing power from overseas facilities to train advanced AI models.

The US Commerce Department's Bureau of Industry and Security (BIS) issued the interim final rule (IFR) on 16 November. It blocks exports, re-exports and in-country transfers of high-end chips to data centres globally if operators know the hardware will supply substantial compute to companies in Macau or Country Group D:5 nations, which include China. The rule is designed to close a specific gap: Chinese firms could previously sidestep chip export controls by renting compute remotely rather than buying hardware.

The threshold is concrete. The rule applies to clusters above 12 million aggregate total processing performance, whether the capacity sits on one site or is spread across linked facilities. Data centre operators, chip exporters and foreign foundries must now verify who ultimately owns their customers, a due-diligence requirement aimed at preventing evasion through corporate spinoffs.

What does the rule change for operators?

Sites already serving restricted firms with high-volume remote compute have 90 days to secure a BIS licence or face export blocks. That is a three-month compliance window for a market where advanced AI training capacity is concentrated in a small number of hyperscale players.

Not every workload falls within scope. Commercial recommender systems at online retailers and social platforms are exempt, provided operators file compliance disclosures and the tools do not generate new content. The carve-out matters for e-commerce and social platforms running large inference and ranking clusters that would otherwise trip the 12 million processing performance threshold.

Which countries face new licensing?

Commerce is also requiring licences for advanced chips shipped to Indonesia, Malaysia, Singapore, Thailand and Brazil, in a bid to stop chips being diverted through third countries. The five markets are major transit and assembly points in the global semiconductor supply chain, and two of them, Singapore and Malaysia, sit at the centre of recent diversion allegations.

US-headquartered companies and their subsidiaries in those countries are exempt, though exporters must file monthly transaction reports with BIS. That reporting requirement gives Washington continuous visibility into chip flows through the region rather than a one-time approval.

The Commerce Department stated its intent plainly: "The rule is designed to ensure that existing restrictions on access to sensitive American technology cannot be circumvented through remote computing arrangements involving data centres in third countries."

The IFR also scrapped the AI Diffusion Rule from the administration of previous President Joe Biden, removing its country tiers, data centre validated end-user authorisations and controls on AI model weights. The rollback replaces a tiered global framework with a targeted enforcement model focused on ownership checks and licence conditions.

Why did the rule land now?

The timing is difficult to separate from the Nvidia leadership crisis. Jensen Huang resigned as president, chair and CEO of Nvidia on 12 November, with CFO Colette Kress named interim chief. The resignation came days after the company opened a probe into his admitted affair with a woman the US accuses of working for Chinese intelligence.

The woman, Alice Huang, is a former executive at Megaspeed, a Singapore-based cloud firm and Nvidia partner that is under US and Singapore scrutiny over alleged diversion of restricted chips to China, the Associated Press reported. Megaspeed has denied wrongdoing. The company told the AP it runs a "legitimate and lawful business" and stated Alice Huang left on 30 August 2025 and holds no current role.

Reuters reported the rule arrived as President Donald Trump departs for an Asia trip that includes a meeting with Chinese President Xi Jinping. The news agency also noted Commerce Secretary Howard Lutnick faces growing criticism over his handling of US export controls on advanced chips to China.

For data centre operators outside the D:5 group, the practical task is immediate: map customer ownership structures, audit any clusters above the 12 million threshold, and file for BIS licences within 90 days, with monthly reporting obligations to follow for flows through the five newly licensed markets.

Also reported

Share this article:

« Previous

More from Daniel Okafor

Daniel Okafor

Show full bio

Senior reporter covering regulation, policy and their impact on Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.

546 articles