# AI Chip Export Controls Now Reach Rented GPUs, and Malaysia and Singapore Need a License

> Source: <https://hwbusters.com/news/ai-chip-export-controls-now-reach-rented-gpus-and-malaysia-and-singapore-need-a-license/>
> Published: 2026-11-17 10:10:44+00:00

**Washington has moved to close the cloud loophole: new AI chip export controls require licenses for data centers anywhere in the world that could sell serious GPU time to Chinese-linked customers, and five countries, from Malaysia to Brazil, lose license-free access to advanced accelerators.**

The US Commerce Department’s Bureau of Industry and Security (BIS) rolled out an interim final rule on Monday that extends AI chip export controls from the hardware itself to the compute it produces. Until now, a Chinese AI lab that could not buy Nvidia accelerators could often rent them instead, from a data center in Singapore or Johor that bought the cards legally. That route is now meant to close.

According to [TechCrunch](https://techcrunch.com/2026/11/16/us-tightens-ai-chip-export-controls-to-deter-chinese-companies-from-accessing-compute-remotely/), a license is now required to export, reexport or transfer AI chips and servers to any data center, in any country, if the exporter may end up providing compute to companies headquartered in the D:5 group of arms-embargoed countries. That list covers China, Russia, Iran, North Korea and 18 others. The rule describes itself as “a logical extension of existing controls,” which is a fair reading: shipping the silicon to those buyers already needed a license.

## Where the line sits

The trigger is capacity, not chip count. The rule applies to data centers or networks that can supply more than 12 million TPP (total processing performance, the metric US export rules use to grade accelerators). TechCrunch puts that at roughly the peak output of 750 H100 or H200 GPUs, a modest cluster by 2026 standards and well within reach of the regional GPU clouds that have sprung up around Southeast Asia.

A few details matter for anyone running hardware at that scale:

- Renting through brokers or intermediaries counts. Exporters and operators are expected to check who actually owns and controls their customers.
- Commercial recommendation systems, the kind that power e-commerce and social feeds, are carved out.
- Operators already serving affected customers have 90 days from the rule taking effect to obtain licenses.

TechCrunch also reports that the rule formally rescinds the Biden-era AI Diffusion Rule of January 2025, which carved the world into tiers with country-level caps. The administration had said in May that it would scrap that framework, and this is the replacement: targeted at access paths rather than at whole countries.

## Five new license countries

Alongside the remote-access piece, BIS added license requirements for advanced semiconductors headed to Malaysia, Singapore, Indonesia, Thailand and Brazil, which Commerce says present a “higher-than-normal risk of diversion.” [Free Malaysia Today](https://www.freemalaysiatoday.com/category/highlight/2026/11/17/us-curbs-export-of-advanced-ai-chips-to-malaysia-singapore) reports that US companies running data centers in those countries are not covered by the new requirement, so the hyperscalers building out Johor should be able to keep importing for their own operations. Locally owned operators are the ones facing new paperwork.

Malaysia is the obvious pressure point. The same outlet reports that Malaysian authorities are investigating Speedmatrix, a tenant at local data centers linked to Singapore-based cloud provider Megaspeed, over US allegations that Nvidia chips were diverted to China. Megaspeed has denied sending chips to China, [Euronews](https://www.euronews.com/2026/11/16/us-clamps-down-on-chip-loophole-in-foreign-access-to-advanced-ai-chips) reports. The company has also been at the center of the scandal that ended with Jensen Huang stepping down as Nvidia CEO last week.

## What it does not settle

The sources do not quite agree on the exemption. TechCrunch describes it as chips destined for US-based companies, while FMT frames it around US firms operating data centers in the five countries, and those are not the same test. Enforcement is the other open question. Checking the beneficial owner of every customer behind a reseller is easy to require and hard to audit, and a 90-day window is short for operators who have never had to do it.

For PC builders the direct effect is close to nil, since the threshold is set at cluster scale, not at a gaming card. The bigger picture is that Washington now treats GPU time as an export in its own right. The timing is pointed too: the rule lands days before President Trump travels to China for a summit with Xi Jinping, and in the same week Nvidia reports its quarterly results.
