Export controls become an alliance issue

For three years, the fight over artificial intelligence export controls looked like a United States-versus-China story. In April 2026, Atlantic Council researchers argued it has become a United States-and-Europe story too, because the argument is now about allied standards, loopholes, and who gets access to the same computing stack. (atlanticcouncil.org) That shift happened because Washington stopped treating controls as a narrow ban on a few chip sales. The Bureau of Industry and Security said on January 15, 2025 that its rule covered advanced computing chips and, for the first time, some artificial intelligence model weights, which are the saved parameters that make a trained model usable. (federalregister.gov) The same January 2025 rule also tried to sort countries by trust level. It created license exceptions and a special data-center authorization for destinations that the Commerce Department said did not raise national-security or foreign-policy concerns. (federalregister.gov) That country-ranking logic hit a political wall fast. On May 13, 2025, the Commerce Department rescinded the Biden-era Artificial Intelligence Diffusion Rule and said it had risked “downgrading” dozens of countries to second-tier status, while promising a replacement approach built around trusted foreign partners. (bis.gov) Europe was not standing still while Washington rewrote the map. On September 8, 2025, the European Commission updated the European Union dual-use control list to add new controls on advanced computing integrated circuits, semiconductor equipment, and related technologies across all member states. (policy.trade.ec.europa.eu) That matters because export controls now reach much further than the box with the chip in it. The Oxford Global Cyber Security Capacity Centre said on March 27, 2026 that the artificial intelligence supply chain includes hardware, compute infrastructure, data flows, and deployment environments, which means a policy change can ripple from procurement to model deployment. (gcscc.ox.ac.uk) Companies have already been redesigning products around those lines. After the United States imposed new licensing requirements on Nvidia’s H20 chip in April 2025, Reuters reported in May 2025 that Nvidia planned a downgraded China version for release in July, showing how export rules can rewrite a vendor roadmap, not just block one shipment. (cnbc.com) That is why the transatlantic argument is now about enforcement gaps. The Atlantic Council brief published on April 9, 2026 warned that without more structured cooperation, the United States and the European Union could drift apart on artificial intelligence policy even while both sides tighten national-security rules. (atlanticcouncil.org) So the practical question inside companies is no longer just “is this legal to export.” It is “which customer qualifies, which region gets which product, which cloud build can be deployed there, and which supplier creates a diversion risk,” because the Bureau of Industry and Security said on May 13, 2025 that it was also issuing guidance on overseas chip diversion and on Chinese model training risks. (bis.gov) The result is that export policy now acts more like an operating condition than a compliance memo. When Washington controls chips, Brussels updates dual-use lists, and security researchers map hidden dependencies across data, models, and infrastructure, procurement teams end up making national-security decisions one purchase order at a time. (policy.trade.ec.europa.eu) (gcscc.ox.ac.uk)