The US government is drawing up sweeping regulations that would grant Washington the authority to approve or block the global expansion of AI infrastructure. In a move that could alter the economics and geopolitics of data centre development for years to come.

Officials at the US Commerce Department have drafted rules that would require companies anywhere in the world to obtain American government approval before purchasing AI accelerator chips from manufacturers including Nvidia and AMD. The proposal, represents a dramatic expansion of existing export controls that currently apply to roughly 40 countries, extending the licensing requirement to virtually every nation on earth.

The draft regulations remain subject to revision and could yet be shelved entirely, but their existence signals the extent to which the Trump administration views AI chip supply chains as a strategic lever, and it is one that it intends to pull.

A licence regime for the AI era

Under the proposed framework, the approval process would be tiered according to the scale of a company’s procurement ambitions. Purchases of up to 1,000 of Nvidia’s latest GB300 graphics processing units would face a relatively streamlined review. Organisations seeking to build larger clusters would be required to obtain pre-clearance before even applying for an export licence.

Those conditions could include requirements to disclose business models in detail, accept US government site visits to data centre facilities, or meet other criteria tied to the specific nature of the deployment, in effect, a diplomatic one.

For the data centre industry, which has spent the past two years in a frantic race to secure GPU capacity, the prospect of a federal licensing layer introduces a new and unpredictable variable into capital planning. Lead times for high-end AI compute are already measured in months; regulatory review cycles have the potential to extend those timelines further still.

Markets responded quickly as Nvidia shares fell as much as 1.9 per cent on the news, while AMD declined 2.3 per cent, both hitting session lows.

Washington: The AI infrastructure gatekeeper?

Officials are keen to stress that this is not an export ban; rather, the Trump administration aims to make American AI technology the global benchmark. However, the approach would effectively give the US government influence over major AI infrastructure investments worldwide.

That is a remarkable position for any single government to occupy. Nvidia’s GPUs are not merely components; they are, for the time being, the irreplaceable foundation of modern AI compute. OpenAI, Google DeepMind, and virtually every other organisation running frontier AI workloads depend on them.

The geopolitical ramifications are immense. Governments in Europe, the Gulf, South-East Asia and beyond have committed hundreds of billions of dollars to sovereign AI strategies, and a significant proportion of that spending is premised on access to Nvidia hardware.

Several of those programmes are being built explicitly to reduce dependence on foreign technology platforms, a goal that sits in uncomfortable tension with a framework that would require them to seek Washington’s approval before breaking ground.

The rule would also raise pointed questions about competitive neutrality. American hyperscalers, such as AWS, Microsoft Azure, Google Cloud, operate data centres globally and would be subject to the same licensing requirements as their international counterparts.

An acceleration of AI export control

It’s not a new concept; the Biden administration spent its final years in office tightening export controls on advanced semiconductors, which culminated in the so-called AI diffusion rule issued in January 2025, which introduced a three-tier country classification system for chip exports.

The Trump administration initially signalled it might roll back some of those controls as part of a broader deregulatory agenda. However, the Commerce Department’s current draft suggests a different calculation has taken hold.

The shift reflects a growing consensus within Washington that dominance in AI is inseparable from control over the hardware that makes AI possible; chips have become what oil was in the 12th century.

The regulatory environment surrounding AI compute is tightening rather than loosening, and there is little sign it will stabilise in the near future. The draft rule may change substantially before it is finalised, or it may never reach that stage at all. Yet its very existence is significant. It introduces uncertainty around deployment timelines, competitive neutrality, and most critically, digital sovereignty.