---
title: "Europe's Sovereign AI Push: Plenty of Talk, Not Much to Show"
description: Anthropic's halt to European access revived the call for homegrown AI. The funding, the labs and the chips show how far Europe's sovereign AI still has to go.
author: Darie Nani (Editor-in-Chief)
date: 2026-06-15T09:13:52.949Z
updated: 2026-06-15T09:13:52.957Z
canonical: https://www.sovereignmagazine.com/article/europe-sovereign-ai-talk-vs-action
image: https://cdn.nanimediahouse.com/europe-sovereign-ai-featured.webp
categories: Artificial Intelligence, Science & Tech, Business
content_type: Analysis
publication: Sovereign Magazine
---

When Anthropic disabled its Fable 5 and Mythos 5 models this month, complying with a Trump administration directive that cited national security, the response from European capitals was immediate and familiar. The order targeted non-US nationals, but because the company had no way to verify the citizenship of its users, it switched the models off for everyone. As Anthropic put it, "the net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance." A nation dependent on others for its technology, said Bruno Retailleau, the French presidential candidate, "is a nation that can be unplugged overnight." Benjamin Haddad, a French minister, argued that Europe "cannot settle for being an open market dependent on technologies designed, funded, and controlled elsewhere." Alec Carns, a British MP, called it "the story of every industry we used to lead."

The sentiment is not new. European leaders have spent more than a year describing artificial intelligence as a test of the continent's independence. The harder question, the one the speeches tend to skip, is whether the money, the companies and the infrastructure behind the rhetoric actually exist. On the available evidence, the gap between what is promised and what has been built remains wide.

## European AI Companies Still Sit a Tier Below

The most-cited proof that Europe can compete is [Mistral AI](https://www.sovereignmagazine.com/article/mistral-ai-and-europe-s-push-for-autonomous-ai-systems), the Paris-based lab founded in 2023. It is a genuine business. Its chief executive, Arthur Mensch, told the Financial Times in February that annualised revenue had passed $400m, growing roughly twentyfold in a year, and the company raised €1.7bn in September at an €11.7bn valuation, with the Dutch chip-equipment maker ASML taking around an 11% stake. A further round at a reported €20bn valuation is said to be in discussion but has not closed.

That is the high point. Set against the American leaders, the scale is different by an order of magnitude. Anthropic raised $30bn in February at a $380bn valuation, roughly seventeen times Mistral's largest round to date. No European model currently appears in the top benchmark tier occupied by Anthropic's Claude, OpenAI's GPT line and Google's Gemini, and [Chinese open-weight labs](https://www.sovereignmagazine.com/article/why-europe-is-caught-between-american-and-chinese-tech-giants) such as DeepSeek and Alibaba's Qwen now rank above Europe's best on public leaderboards.

Behind Mistral the field thins quickly. Aleph Alpha, once Germany's national champion, stopped training frontier foundation models in 2024 and repositioned as a software layer for regulated industries; it is now being absorbed by the Canadian firm Cohere in a deal announced in April. Black Forest Labs, the other German name often mentioned, is funded largely by American venture capital. The roster of independent European frontier labs, in practice, is short.

## Europe's AI Investment Numbers Are Smaller Than They Sound

The figure attached to Europe's ambition is €200bn, announced by Commission President Ursula von der Leyen at the Paris AI summit in February 2025 under the InvestAI banner. It is not €200bn of new public money. The sum combines an industry-led pledge of about €150bn with roughly €50bn drawn from existing EU programmes, and no official line-by-line breakdown has been published.

The part that is genuinely ring-fenced is much smaller: €20bn for "AI gigafactories," large compute sites each meant to house 100,000 or more advanced chips. Here the detail matters. Under the governing regulation the EU budget covers only around 17% of each site's capital cost, leaving roughly two-thirds to be found from private investors. Bloomberg reported on 2 June that the programme is "floundering," with the formal call slipping to around July 2026, funds earmarked only for 2028 and 2030, and corporate interest narrowing from about 70 companies to roughly 10.

A smaller, older scheme has fared better. The EuroHPC programme has designated 19 "AI factories" attached to public supercomputers, and JUPITER, Europe's first exascale machine, went live at Jülich in Germany in 2025 and was inaugurated by Chancellor Friedrich Merz in November. It is the one piece that has unambiguously been built. It also runs on around 24,000 Nvidia superchips.

## Sovereign AI That Runs on American Chips

That last detail points to the central contradiction. Almost every "sovereign" build Europe has announced depends on American silicon and, often, American-owned infrastructure. Mistral's planned data centre near Paris is specified around Nvidia hardware. Deutsche Telekom's "industrial AI cloud" in Munich, opened this year, uses Nvidia chips. Even Amazon Web Services markets a "European Sovereign Cloud," launched in Brandenburg in January, which is sovereign in operation but American in ownership.

Europe is not absent from the [chip supply chain](https://www.sovereignmagazine.com/article/european-semiconductor-companies-chips-act). ASML, based in the Netherlands, holds close to a complete monopoly on the extreme ultraviolet lithography machines without which no advanced chip in the world can be made. But there is no European designer of frontier AI processors to rival Nvidia, and the continent's flagship processor effort, SiPearl, supplies a CPU that sits alongside Nvidia's GPUs rather than replacing them.

The cloud picture is similar. The American "big three," AWS, Microsoft Azure and Google, hold around 70% of the European cloud infrastructure market, according to Synergy Research, while domestic providers have stayed flat at roughly 15% for years. OVHcloud, the largest European player, passed €1bn in revenue for the first time in its last financial year but reported net income of just €0.4m. Gaia-X, the project launched in 2020 to build a European cloud federation, is now widely cited as a cautionary tale; one founding member quit in 2021, accusing it of "sovereignty-washing."

## Digital Sovereignty Meets the Regulation Reversal

The policy signals have also pointed in two directions at once. In July 2025 a Commission spokesman insisted there would be "no stop the clock" on the EU AI Act and "no pause." Four months later the Commission itself tabled the Digital Omnibus, a package that proposed exactly such a delay. Under a provisional agreement reached in May, the Act's high-risk obligations have been pushed back to December 2027.

The reversal followed sustained pressure from European industry. More than 45 companies, including ASML, Airbus, Mercedes-Benz and Mistral, signed a "Stop the Clock" letter in July 2025 asking for a two-year pause on the rules. A separate open letter from over 100 firms had earlier demanded a "[buy European](https://www.sovereignmagazine.com/article/european-digital-stack-can-europe-build-its-own-eurostack-for-digital-sovereignty)" approach to public technology procurement. Both reflect a recurring complaint, set out most prominently in Mario Draghi's 2024 report on European competitiveness, that the continent regulates new technology before it has built any.

Draghi's diagnosis was blunt: he estimated that Europe needed more than €800bn in additional annual investment to close its productivity gap with the United States, and noted that the bloc was producing a handful of AI foundation models a year against dozens in America. Eighteen months on, the headline response has been a deregulation package and delayed deadlines rather than investment on anything like that scale.

## What Has Actually Changed

The Anthropic episode will be read in Brussels and Paris as confirmation of a thesis European leaders already hold: that depending on foreign technology is a strategic risk. That thesis is reasonable. The difficulty is that the practical answer to it, a sovereign stack of European chips, clouds, models and capital, does not yet exist in any form that could absorb the loss of access to an American provider.

There are real assets to build on. Mistral is a credible lab, ASML is irreplaceable, and JUPITER works. But a single frontier company sitting a generation behind on roughly a seventeenth of the capital, a compute programme reported to be stalling before its money arrives, and an infrastructure base still running on American hardware do not add up to independence. For now, Europe's sovereign AI remains more a statement of intent than a working alternative.

## FAQ

**Q: Is there a European AI model like ChatGPT?**
Yes. Mistral AI in France builds large language models and a consumer assistant, Le Chat, and reported annualised revenue above $400m in early 2026. Its models are competitive but sit a tier below the leading systems from Anthropic, OpenAI and Google on public benchmarks.

**Q: What are the EU AI gigafactories?**
They are large compute sites, each intended to house 100,000 or more advanced AI chips, backed by about €20bn of EU and member-state support. As of mid-2026 the formal funding call had slipped to around July 2026, with money earmarked for 2028 and 2030, and the programme has been reported as struggling to attract private investors.

**Q: Why does Europe depend on the United States for AI?**
Europe lacks a domestic designer of frontier AI processors, so its data centres run on American Nvidia chips, and US providers hold around 70% of the European cloud market. The continent's strength lies upstream, in ASML's near-monopoly on the lithography machines used to make advanced chips.

**Q: Which countries lead in sovereign AI?**
The United States and China lead on frontier models, capital and compute. Europe has one significant independent lab in Mistral and strong positions in chip-making equipment, but no homegrown frontier processor or large-scale cloud to match the American hyperscalers.