Europe courts Anthropic, but the numbers don't add up: why the Austrian offer reveals a deeper structural gap

By Momentum IA · June 28, 2026.

On June 12, 2026, the U.S. Department of Commerce issued an export-control directive prohibiting any foreign national —including Anthropic's own non-U.S. staff— from accessing the company's two most powerful models: Claude Fable 5 and Claude Mythos 5, launched just days earlier. The measure was not arbitrary: researchers at Amazon, Anthropic's largest investor, had detected that Mythos 5 had provided restricted guidance on cyberattacks. CEO Dario Amodei described the incident as a «narrow bypass, not a full jailbreak», but the model had already demonstrated the ability to compromise protected government systems. Unable to filter users by nationality, Anthropic withdrew both models globally. Only Claude Opus 4.8 remained operational. On June 26, Washington partially eased the block for more than 100 trusted U.S. institutions; Fable 5 remains restricted.

Against this backdrop, Alexander Pröll, Austria's Secretary of State for Digitalization, sent a letter to the European Commission's Executive Vice President, Henna Virkkunen, asking that Europe explore «the strategic establishment and participation of Anthropic within the European Union». The incentives on offer —legal certainty, fresh capital and access to the single market— sound attractive on paper. The problem is that Pröll himself admitted in the same letter that he did not know how the plan would work in practice. There is no funding figure, no timeline, no technical roadmap.

**The arithmetic that exposes the proposal**

Anthropic is not a relocatable startup. The company is financing the construction of a $50 billion data center spread across Texas and New York. Amazon has invested $13 billion and is its main training partner. In return, Anthropic has committed to spending more than $100 billion on Amazon's cloud over a decade. Moreover, the company itself estimates that U.S. AI will need around 50 gigawatts of additional electrical power by 2028. Europe, as a whole, has neither that infrastructure nor anything close to that time horizon.

The contrast with European ambitions in semiconductors illustrates the problem. The European Chips Act aspires for Europe to account for 20% of global production by 2030, up from less than 10% today. The Commission's own forecasts point to 11.7%, and European auditors call the original target «highly unlikely». The gap between political rhetoric and the material base is exactly the same kind of gap that separates the Austrian offer from a viable proposal.

**Our read: a political signal with real consequences**

It would be easy to dismiss this episode as an empty gesture. That would be a mistake. What the Fable 5 and Mythos 5 incident reveals —and what Pröll's letter articulates in a clumsy but sincere way— is a structural dependency that Europe has been ignoring for years: its administrations, its companies and its researchers access frontier AI through American platforms subject to U.S. national-security legislation. A single executive order can cut off that access globally, without warning, without the possibility of immediate appeal.

That is not a public-relations problem. It is a digital-sovereignty problem that goes far beyond Anthropic. Today's restriction affects two specific models over a specific security incident. The pertinent question is what happens when geopolitical tensions intensify or when the American national-security framework broadens. Europe needs answers to that question that do not depend on persuading a private American company to move its $50 billion infrastructure across the Atlantic.

The genuine answer involves investing massively in homegrown models —initiatives such as Mistral already exist— and building the compute, energy and capital infrastructure that makes the U.S. the natural home of frontier AI. That is a project of decades, not of a legislative term. In the short term, Europe will face an uncomfortable transition in which its access to the most advanced capabilities will be subject to decisions made in Washington, not in Brussels. The illusion that a well-intentioned letter is enough merely postpones harder conversations about real investment, industrial policy and appetite for technological risk.

What Europe can do immediately is what it already does relatively well: regulate coherently and create legal certainty. But regulation without production capacity is arbitrage, not sovereignty. The long term is hopeful —Europe has talent, sophisticated capital and an internal market of 450 million people—, but reaching that long term with real autonomy demands investment decisions that today still go unmade.