Austria’s proposal to bring Anthropic inside the EU stopped being a thought experiment on July 15, 2026, when Brussels formally published the Cloud and AI Development Act in the Official Journal, the law meant to make AI sovereignty something regulators can actually measure and enforce rather than a talking point in a letter. The timing is not coincidental. Austria’s State Secretary for Digitalization, Alexander Pröll, wrote to European Commission Executive Vice President Henna Virkkunen on June 28 asking member states to jointly explore hosting Anthropic within EU borders, less than three weeks before the legal framework that would actually classify such a move went on the books.
Pröll’s letter followed a June 12 US export-control order that cut off access to Anthropic’s Fable 5 and Mythos 5 models for every user outside the United States, an episode Anthropic itself has confirmed and that LiveAIWire covered in detail in our report on the precedent the shutdown set for AI export controls. Pröll’s letter, reported by Reuters, argued that only relocating the legal jurisdiction controlling a model’s access, not the company’s headquarters, would put a future kill switch out of Washington’s reach.
Why AI Sovereignty Now Has a Legal Definition
The Cloud and AI Development Act gives Brussels something it did not have when Pröll wrote his letter: a formal, tiered definition of what counts as sovereign AI infrastructure. According to the European Commission’s own summary of the regulation, the law establishes four assurance levels that public sector bodies can require of cloud and AI providers as part of the EU’s broader AI sovereignty framework. Level one requires only that data be processed and stored inside the Union. Level two requires demonstrated independence from third countries and transparency over the software supply chain.
Level three requires that a provider actually be owned and controlled from the EU, with additional criteria including personnel citizenship, though the Commission can grant exceptions for third-country providers. Level four requires full transparency and control over the software supply chain with no third-country interference at all.
That tiered structure is what makes Austria’s proposal so much harder than it sounds. Hosting Anthropic within the EU in a way that satisfies AI sovereignty at level three or four would mean genuine EU ownership and control, not a European subsidiary that still answers to a US parent company and US export-control law. Pröll’s own letter acknowledged this directly, offering legal certainty, single-market access, and capital as incentives without specifying how a relocation of that depth could actually happen.
What Anthropic Would Have to Give Up
The practical barriers to a level-three or level-four AI sovereignty move are substantial regardless of what the new law defines. Anthropic has committed roughly fifty billion dollars to US computing infrastructure, including data centres built with Fluidstack in Texas and New York, and has committed more than one hundred billion dollars over the next decade to Amazon Web Services technologies, securing up to five gigawatts of new compute capacity through that partnership. Amazon has invested thirteen billion dollars in Anthropic directly and remains its primary cloud infrastructure partner. Anthropic is simultaneously preparing a US stock market listing. None of that capital, compute, or corporate structure moves to Europe because a letter from Vienna asks it to.
Europe’s own capacity constraints compound the AI sovereignty problem. The Cloud and AI Development Act’s stated goal is to at least triple the EU’s data centre capacity within five to seven years, which is itself an admission that the bloc does not currently have anywhere near the compute base needed to host a frontier AI lab at the scale Anthropic already operates in the United States. AI sovereignty on paper and AI sovereignty in gigawatts are different things, and the gap between the two is measured in years, not months.
The Pattern This Fits Into
Austria’s letter followed a wider pattern of US government leverage over AI infrastructure that has alarmed European capitals well beyond this single episode. The Five Eyes joint cybersecurity warning issued in June 2026 argued that frontier AI capabilities were becoming critical national security infrastructure days before those same capabilities became subject to a unilateral US export-control decision that European governments had no role in shaping. The dynamic echoes what happened earlier with Nvidia’s chip export restrictions aimed at China, except this time the restricted party was Washington’s closest allies rather than a strategic rival.
The Cloud and AI Development Act’s own explanatory materials describe over reliance on non-EU cloud providers as posing a significant risk to Europe’s digital autonomy and resilience, language that reads, in the wake of the Anthropic shutdown, less like abstract policy prose and more like a direct response to what June 12 demonstrated in practice.
What This Means for You
If your organisation operates in the EU and depends on US-hosted frontier AI models, the Cloud and AI Development Act is the first concrete signal of how Brussels intends to price that dependence going forward. Public sector bodies and, eventually, operators of critical infrastructure will be expected to assess providers against the four-level framework, and providers unable or unwilling to meet the higher tiers may find themselves locked out of a growing share of public procurement even if their commercial terms remain otherwise unchanged. The rollout is gradual, with the highest tiers not becoming mandatory for several years, but the direction of travel is now written into law rather than left to diplomatic letters.
For AI sovereignty to become more than a legal category, Europe needs the underlying compute, capital, and industrial base to make level-three and level-four providers commercially viable, not just legally defined. Austria’s proposal named the ambition. The Cloud and AI Development Act gives that ambition its first real yardstick. Whether either produces a European frontier lab operating outside Washington’s reach, or simply a more expensive way of measuring a dependence that persists regardless, is the question the next several years of implementation will answer.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and geopolitics. LiveAIWire publishes daily AI news and analysis at liveaiwire.com.