VMTech
Discuss a project →

US plans licences for AI compute that can be accessed remotely

US plans licences for AI compute that can be accessed remotely

The US Commerce Department is preparing an interim export-control rule that would require licences for the export, reexport or transfer of AI chips and servers to data centres anywhere in the world when those facilities may supply compute to companies in China or other restricted countries. The measure is intended to stop companies from bypassing existing US chip controls through remote access to infrastructure located abroad.

The rule targets organisations based in the US D:5 group of embargoed countries, which includes China, Russia, Iran, North Korea and 18 other jurisdictions. Existing restrictions already require chip companies to obtain licences before shipping AI chips to entities based in, or owned by businesses in, these countries. The proposed remote-access provisions extend that scrutiny to the compute services enabled by overseas data centres.

Compute threshold and customer checks

The requirements would apply to data centres and networks able to provide more than 12 million total processing performance, or TPP, the export-control metric used by the US to measure computing power. The source estimates that level as roughly equivalent to the maximum computing power of about 750 Nvidia H100 or H200 chips.

The interim rule would also cover indirect access arranged through intermediaries or brokers. Exporters and data-centre operators would be required to monitor customers' ownership structures and activities to establish whether facilities are controlled by, or provide compute to, parties from D:5 countries.

The Commerce Department said the restrictions are a logical extension of existing controls and are intended to limit the ability of companies headquartered in countries of concern to use US advanced Super Intelligence semiconductors to train advanced Super Intelligence models. It said access to that level of computing power can support training and operation of models with potential military- and intelligence-related uses that could threaten US national-security or foreign-policy interests.

Additional licensing for selected markets

The rule would add licensing requirements for exports, reexports and transfers of advanced semiconductors to Malaysia, Singapore, Indonesia, Thailand and Brazil. The department identified those destinations as presenting a higher-than-normal risk of diversion. The additional requirements would not apply to chips destined for companies based in the United States.

Existing data-centre operators delivering services covered by the rule would have 90 days from its effective date to obtain licences. Commercial recommendation systems, including those used in e-commerce and social media, are excluded from the rule.

Replacement for the AI Diffusion Rule

The proposal would rescind the Biden administration's January 2025 AI Diffusion Rule, which sought country-level limits on exports of advanced AI chips and restrictions on transfers of certain powerful AI models. The Trump administration said in May that it would discard that framework, arguing it imposed unnecessary limits on US companies and allies.

For businesses that export advanced chips, operate international data centres or procure AI capacity through third parties, the immediate implication is to map compute capacity, customer ownership and broker relationships against the new licensing triggers before the 90-day compliance window begins.

#aichips#exportcontrols#datacenters#compliance
Open analytics
On the site 3 views
min read 4 16.11.2026
Instagram

US plans licences for AI compute that can be accessed remotely

Open the post on Instagram ↗