Series: What Europe’s AI Market Actually Looks Like

This series is published in partnership with the AI Now Institute.

Europe, Spring 2026. Germany’s AI translation darling DeepL partners with Amazon Web Services. French Mistral develops their models in partnership with Nvidia. Swedish Lovable gets a jump scare as Anthropic moves in to eat its lunch. Meanwhile, across Europe, companies who submitted bids to build the continent's ‘Gigafactories’ — large-scale computing infrastructure designed to help Europe train the latest generation of frontier AI models — are threatening to pull the plug.

For sovereignty-minded EU politicians, this landscape might cause distress. Massive amounts of political energy are being spent to break free from US Big Tech: by deregulating, providing capital, aspiring to build sovereign computing clusters, and nurturing ambitious tech startups. Yet, the initiatives keep being frustrated by the gravity wells of the global tech market that bend the interests of private actors in their direction.

Over the past two years, we have systematically tracked the growing political effort to build "Sovereign AI” in Europe. A key theme throughout this work has been that Europe’s AI industrial policy rests on an implicit, largely unexamined assumption: that European AI companies, their funders, and their market incentives are naturally aligned with European strategic interests. Simply put, the more European AI companies there are, the better for Europe.

The reality is much messier. What looks like a European AI ecosystem is a confluence of interests pulling in different directions: VC timelines that reward fast exits over long-term independence, market structures that funnel revenue upstream to US providers, infrastructure dependencies that deepen with every investment, and playbooks borrowed wholesale from Silicon Valley.

The policy problem runs even deeper. This week, the European Commission presented the European Technological Sovereignty Package, a set of measures to strengthen Europe's capacity in semiconductors, AI, cloud and open source. This is more than a framing exercise. With the proposed Chips Act 2.0 and the Cloud and AI Development Act, the package contains two concrete legislative proposals that are deeply revealing about the unspoken assumptions that drive Europe’s strategy on AI: about where value is created and captured in AI supply chains, about AI's future trajectory, and about whether sovereignty can be achieved without engaging in market shaping.

Simply put, Europe’s strategy on AI has focussed on two sides of the market: boosting demand (via initiatives like the Apply AI Strategy, aimed at pushing AI adoption), and — primarily — increasing supply of inputs (via the Cloud and AI Development Act, expanding on compute and data center capacity and the proposed Cloud Act 2.0, aimed at increasing Europe’s chips supply). What was initially framed under the umbrella of competitiveness, now includes added measures aimed at ensuring “sovereignty,” with the laudable goal of progressing towards a full European technology stack.

As we will argue throughout this series, these industrial policy proposals rest on an incomplete, two-dimensional view of the European AI market. In reality, this market is deeply entangled with the ecosystems of dominant US players, in ways that boosting both supply and demand alone cannot disentangle. As a result, interventions that seek to secure sovereignty in AI risk leading to much more entrenched dependence.

Most EU tech policy commentary operates top-down: it starts from strategy documents and asks whether targets are being met. This series works bottom-up. It looks at what the European AI market actually looks like (who supplies whom, where revenue flows, what dependencies are baked in) and tests the assumptions that policy documents leave implicit. Where evidence is incomplete or trajectories genuinely uncertain, we will say so openly rather than defaulting to confident prescriptions.

This mini-series consists of three parts: The first piece will examine Europe's AI application layer; how startups like Lovable function as a distribution channel for US models, and what that means for the value Europe actually captures. The second will trace why even Europe's strongest independent AI companies, like DeepL, end up dependent on US hyperscaler infrastructure. The third will pull the threads together: the common structural dynamic, the strategic choices Europe is avoiding, and the honest uncertainties about where AI is heading. These pieces will be published over the coming weeks in Tech Policy Press and on the EU AI Industrial Policy Monitor Newsletter.

The obvious risk is that the widespread adoption of AI further entrenches Europe’s dependence on a handful of US firms. The less obvious risk — and the one this series examines — is that even where Europe builds its own AI capabilities, the economic value still flows elsewhere. Whether and how much to build is itself a political question this series doesn't presume to answer. But whatever Europe decides, an AI industrial policy strategy that cannot decipher its priors — particularly around where and how it will complete — will not be successful. It is a wish list that avoids the scrutiny that comes from clearly naming tradeoffs and hard choices.