Top AI executives are joining G7 leaders in Evian-les-Bains, France, today as the politics of artificial intelligence shift from abstract safety principles toward a harder question: who can keep access to advanced models, compute, cloud infrastructure, and the companies that operate them.
The working lunch, framed around the safe, rapid, and effective deployment of AI, is expected to include OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, Anthropic CEO Dario Amodei, and leaders from smaller AI companies including Cohere, Mistral, Black Forest Labs, Domyn, Sakana AI, and Synthesia, according to Associated Press reporting carried by WTOP. The lineup makes the G7 meeting less like a routine policy session and more like a snapshot of the new AI supply chain: U.S. frontier labs at the center, allied countries trying to avoid dependence, and regional challengers pitching sovereignty as a product feature.
The timing matters. This month, the European Commission proposed a European technological sovereignty package covering semiconductors, AI, cloud, open source, and data-center integration with the energy system. Canada used the recent G7 digital ministers meeting in Paris to emphasize secure and responsible AI, digital-sector resilience, sovereign cloud infrastructure, and AI adoption tied to productivity and economic growth. The United States, meanwhile, has been pushing an AI strategy built around national security, domestic leadership, and exportable U.S. technology stacks.
The Anthropic restriction changed the conversation
The sovereignty debate sharpened after Anthropic suspended access to its Fable 5 and Mythos 5 models following a U.S. government order that barred non-Americans from using them, AP reported. The order cited an unspecified national security concern, and Anthropic’s response effectively removed access for all customers rather than trying to split users by nationality.
That episode turned a policy argument into a procurement risk. A government, bank, hospital, manufacturer, or software vendor outside the United States can no longer treat access to the most advanced U.S. models as a purely commercial contract question. If model availability can change because of national-security action, export controls, cloud restrictions, sanctions, or domestic political pressure, then “AI sovereignty” starts to mean operational continuity as much as national pride.
Axios reported that the move alarmed parts of the AI industry because U.S. labs depend on global adoption of their most advanced models to justify massive infrastructure spending and valuation expectations. For enterprise buyers, the same incident argues for model diversification, portability, and clearer fallback plans before they build critical workflows around a single frontier provider.
Europe wants more than regulation
Europe’s latest sovereignty package is not just another AI Act-style compliance effort. The Commission’s June 3 plan links AI autonomy to chips, cloud capacity, open source, and energy systems. Its proposed Cloud and AI Development Act would support research and innovation, streamline conditions for deploying data centers across the EU, and create a single framework for assessing cloud and AI sovereignty. Chips Act 2.0 is meant to strengthen advanced semiconductor capacity and demand, while the open-source strategy is pitched as a way to reduce lock-in and improve transparency.
That mix shows where the argument has moved. A country cannot have much practical control over AI if its sensitive workloads rely on foreign cloud platforms, imported accelerators, closed hosted models, and data-center capacity whose energy needs are politically fragile. Sovereignty, in this version, is less about banning U.S. AI services than about making sure public agencies and regulated industries have alternatives when access, pricing, model behavior, or policy conditions change.
That is why companies such as Mistral and Cohere matter at the G7 even though they do not match the reach of OpenAI, Google DeepMind, or Anthropic. Mistral gives Europe a local AI champion and an open-weight strategy that appeals to governments and companies that want more control over deployment. Cohere gives Canada a domestic enterprise AI player at a moment when Ottawa is tying AI strategy to public infrastructure, trusted partnerships, and sovereign cloud capacity.
The U.S. advantage is also a vulnerability
The United States still has the clearest lead in frontier models, AI infrastructure spending, cloud depth, and chip access. The White House’s June 2 executive order on advanced AI innovation and security presents that lead as a strategic asset, calling for coordinated action across agencies as advanced AI capabilities become more important to national security.
But the same concentration gives allies a reason to hedge. If U.S. AI providers become part of a national technology-control system, customers abroad may see them less like ordinary software vendors and more like strategic infrastructure that can be interrupted. That does not mean they will abandon U.S. models. It does mean they will ask harder questions about data residency, exportability, local hosting, audit rights, model switching, and whether mission-critical workflows can keep running if a preferred model disappears from the menu.
For enterprises, the practical lesson is immediate. AI architecture should assume that model access is not permanent. Contracts need portability language. Internal systems should separate application logic from a single model provider where possible. Security and compliance teams should know which AI workloads depend on which jurisdictions, cloud regions, APIs, and data flows. Procurement teams should treat sovereign deployment options, open-weight fallbacks, and multi-model routing as resilience features rather than abstract policy preferences.
What to watch after Evian
The G7 is unlikely to settle the sovereignty fight in one lunch. The more important signs will come afterward: whether leaders endorse new language on AI access and infrastructure, whether Europe turns its sovereignty package into funded deployment programs, whether Canada advances sovereign cloud and AI infrastructure plans, and whether U.S. export-control policy becomes a recurring constraint on frontier models.
The broader shift is already visible. AI policy is no longer only about model safety, copyright, bias, or chatbot rules. It is becoming an infrastructure contest over chips, data centers, cloud control, open-source alternatives, and the legal power to decide who gets access to the most capable systems. The companies at the G7 table may build the models, but governments are making clear that model access is now part of industrial strategy.
