U.S. officials are debating a new framework for artificial intelligence chip exports that would tie access to advanced American semiconductors to investment in U.S. data centres or to formal security assurances from foreign governments. According to documents seen by Reuters, the draft would apply to orders of 200,000 chips or more and would mark a significant expansion in how Washington manages the global flow of frontier AI hardware. The proposal is not final and could still change.
Even installations of fewer than 1,000 chips could require a licence. Orders of up to 100,000 chips could require government-to-government assurances, while installations of up to 200,000 chips could trigger visits from U.S. export-control officials. To qualify for certain exemptions, exporters such as Nvidia and AMD would also need to monitor the chips, and recipients would need software restrictions designed to stop them being linked into larger computing clusters.
The draft suggests Washington is moving beyond using export controls simply to restrict access for strategic rivals. Access to U.S.-designed AI chips would also become a lever over where infrastructure is built, where capital is committed, and on what terms foreign buyers are allowed to scale computing capacity. That would pull trade control, industrial strategy, and national security into a far tighter alignment.
The Commerce Department signalled that broader direction in May 2025 when it rescinded the previous AI diffusion framework and said it was committed to promoting secure exports of the American tech stack. The model now under discussion appears to follow that logic — more monitoring, more bilateral oversight, and a closer link between export permission and economic commitments inside the United States.
There are some immediate unanswered questions for businesses. What would count as qualifying investment — equity, data centre construction, energy infrastructure, or a broader package of industrial commitments? Would close partners such as the UK still have a lighter-touch route, or would they also need formal assurances and site-level compliance? And would overseas demand increasingly be channelled through U.S. cloud providers rather than through nationally backed compute projects?
That matters in Britain as government, investors, and infrastructure operators weigh how far the country can build sovereign AI capacity without relying entirely on overseas cloud groups. Across Europe, the Gulf, and Asia, governments are trying to secure access to advanced chips while hyperscalers and investors decide where the next wave of data-centre capacity will sit. If export approvals become tied to domestic U.S. investment, the advantage may shift further towards buyers that can pair chip demand with large-scale capital commitments in America.
Recent U.S. agreements with Gulf states point in that direction. AI cooperation has been linked with commitments to invest in American infrastructure, including data centres and related energy assets. That offers an early indication of how Washington may want this system to work in practice — chip access, strategic alignment, and inward investment moving together rather than as separate negotiations.
For Nvidia, AMD, cloud operators, and governments trying to build independent AI capacity, that would materially change the calculation. The question would no longer be only whether advanced chips can be bought abroad, but on what political, security, and capital terms. Until a final rule is published, businesses are planning around a moving target that increasingly resembles industrial policy delivered through export control.
