arXiv
PubMed
Nature
SEC filings
Financial Times
EUR-Lex
Federal Register
US export controls reshape the AI chip market
Washington has widened the licence requirement for advanced AI accelerators to a second tier of countries, and the three largest chip designers have confirmed they will comply from the next quarter. For buyers outside the first tier the practical effect is a longer lead time, not a ban.
A second tier of countries now needs a licence for top-end accelerators
The new rule extends the licence regime beyond the original list and sets a performance threshold that covers the current generation of datacenter GPUs. NVIDIA and AMD have told customers in the affected countries that existing orders will ship under licence; both expect the review to add weeks, not months, to delivery.
The constraint moves from "can we buy" to "when does it arrive". Capacity planning in the second tier now needs a licence buffer in every timeline.
Cloud providers in the first tier are already selling the gap
Two hyperscalers announced capacity reserved for customers in the newly licensed countries, priced per hour and available this quarter. The pitch is explicit: rent the compute where it is already installed instead of waiting for hardware to clear. In effect, the export rule works as a subsidy for first-tier datacenter build-outs. Ask the brief
Demand that cannot buy chips becomes demand for cloud regions. Watch where the next regions get announced, and by whom.
- ·The first licence decisions: how many weeks the review really takes, per country.
- ·Cloud regions announced in or near the second tier, and which hyperscaler moves first.
- ·Whether the performance threshold shifts with the next GPU generation.
- Financial Times · 2h
- SemiAnalysis · 5h
- NVIDIA Blog · 9h