A coalition of roughly forty economists, AI researchers, and former policymakers published A Transformative AI Strategy for Europe today. The names carry real weight: Margrethe Vestager, the European Commission's former Executive Vice-President for competition policy; Philippe Aghion, who won the 2025 Nobel Memorial Prize in Economic Sciences for work on innovation-driven growth; Daron Acemoglu, the 2024 Nobel laureate whose research has been more skeptical than most economists' about AI's near-term productivity payoff; former Irish Taoiseach Leo Varadkar; and Yoshua Bengio, the Turing Award-winning AI researcher this blog has covered repeatedly this month for his warnings about agents that lie, cheat, and coordinate against oversight. The report states plainly that it wasn't commissioned or funded by the European Commission or any national government — which matters for how much weight to put on its numbers, and also for how much power it actually has to make anything in it happen.
Who's actually behind it
The report is coordinated by the KIRA Center — the Center for AI Risks & Impacts, an independent nonprofit think tank based in Berlin, founded in April 2023 by Daniel Privitera. That's worth naming plainly, because it's not the institutional home a reader would guess from the report's framing. Transformative AI Strategy for Europe reads as an economic-competitiveness document — compute shares, private capital targets, supply-chain chokepoints — but its convening organization's usual focus is AI existential and societal risk, the same territory Bengio has occupied publicly all year. That doesn't make the economics wrong; Aghion and Acemoglu are Nobel-level economists making their own arguments, not KIRA staff reciting talking points. But a safety-focused organization assembling an economic-sovereignty coalition is itself a data point about where the AI-risk community's attention has moved this year: not just "will these systems escape our control," but "who gets to decide the answer to that question if Europe has no seat at the table."
One thing worth stating plainly, because this blog has flagged self-interest in nearly every industry statement it's covered this month: it's hard to find much here. None of the roughly forty authors run a company with a product riding on the report's recommendations. Vestager and Varadkar are out of office. Aghion and Acemoglu are academic economists with no AI lab equity disclosed in the coverage of this launch. Bengio has spent the year warning about the technology this report wants Europe to buy more access to, not selling it. That's a genuinely different posture than Amodei's pacing proposal, Nadella's Code of Conduct, or Mensch's "own your AI" pitch — all of which this blog has noted map suspiciously well onto the speaker's own commercial position. Here, the closest thing to an interest is institutional rather than financial: KIRA's own relevance as a convener, and Vestager's, rests partly on this report landing with European policymakers.
The diagnosis, checked against numbers this blog already has
The report's headline figures are stark, and at least one of them holds up against independent data rather than requiring the report's own word for it. Its claim that the two leading AI labs' combined revenue grew roughly 25-fold in two years, from around 100 billion, is consistent with Epoch AI's independently tracked figures: Anthropic and OpenAI's combined annualized run rate crossed roughly $115 billion by mid-2026, up from a combined figure in the low single-digit billions in 2024. The EU compute-share numbers — roughly 5% for Europe against 75% for the US and 15% for China — aren't independently checkable from here in the time available, but they're the same order of magnitude this blog has seen elsewhere this year in Mistral's own compute-economics coverage, and the report's specific claim that a single Malaysian data-center site under construction will reach roughly a third of Europe's total compute capacity by year's end is plausible on its face: Malaysia has absorbed the majority of Southeast Asia's AI data-center buildout this year, enough that Z.ai's own sovereign-AI partner program already runs an office there.
A recommendation that's already been overtaken by events
The report's most specific piece of advice is also its most interesting to check against what's actually happening: it tells Europe to focus most of its effort on securing guaranteed access to frontier AI built elsewhere, and to attempt a domestic frontier champion only if genuinely demanding conditions — state-scale resource commitment, not "bold announcements without matching resources" — are met. That's a considered, defensible position. It's also a position Europe has already partly overtaken. This blog covered, eleven days ago, France's economy minister Roland Lescure warning in Silicon Valley that "if European AI comes down to just Mistral, we're screwed" — the same underlying anxiety this report now attaches macro figures to. But that same post also covered the EU's actual policy choice, made back in June: the European Commission's Frontier AI Grand Challenge selected the EUROPA consortium, led by Italy's Domyn with Germany's Fraunhofer-Gesellschaft, to build a sovereign 400-billion-parameter open-source frontier model on EuroHPC compute — a real, funded, in-progress attempt at exactly the kind of domestic champion this report says should only be attempted under demanding conditions. Whether EUROPA meets that bar is a fair question this report doesn't appear to engage with directly. And the same week Lescure was in San Francisco warning against over-relying on Mistral, France's finance ministry was reinforcing a €6 million state partnership with Mistral specifically, seconding Mistral engineers directly into government ministries. Both of Europe's live sovereignty bets — an Italian-German consortium model and a deepened French-Mistral partnership — already look more like the "build a champion" path than the "buy access" path this report recommends prioritizing. The report arrives eleven days into a live European policy debate having already lost the argument on the ground, at least so far.
A quieter disagreement about what Europe should actually be exporting
The report's fifth priority — becoming "the global leader in AI assurance technology," the hardware and methods to prove how AI systems are secured and used — sits in interesting tension with a different European argument this blog covered two days ago. Henna Virkkunen, the Commission's tech-sovereignty chief, spent September 12 arguing that Europe should export its AI Act's loss-of-control framework globally, positioning EU law itself as the product other jurisdictions should adopt. This report's version of European leadership is narrower and more technical: not rules for the world to copy, but assurance technology — instruments and methods, sold or licensed, that happen to also serve Europe's own oversight needs. Virkkunen is pitching Brussels as a regulatory standard-setter; this report is pitching Europe as a toolmaker. Both could be true simultaneously, but they're not the same ambition, and a reader tracking both stories in the same week should notice that the EU's own tech-sovereignty chief and today's independent expert coalition are making distinguishable, not identical, cases for what Europe's comparative advantage in AI governance should actually be.
What's genuinely new here
Set against the run of individual reactions and single-company announcements this blog has tracked in the sovereignty debate — a minister's quote, a CEO's tweet, a state's procurement decision — this report's contribution is scale and a number. €100 billion in public investment, targeting €1.5 trillion in private capital, to triple compute share by 2030, is the first attempt in this blog's recent coverage to put a single, EU-wide figure on what "not falling behind" would actually cost, rather than leaving the reader to infer it from a compute-share percentage or a national procurement deal. Whether that number is realistic isn't something this report, or this post, can settle — no independent economic model of the claim is cited in the coverage available, and the authors' own credibility rests partly on Aghion's and Acemoglu's academic standing rather than on a transparent, checkable calculation reproduced here. What the report does credibly establish is the shape of the problem: a currency gap of nearly two orders of magnitude between Europe's AI sector and the two companies it's implicitly competing against, arriving at a moment when France's own government is visibly unable to decide, in public, whether its national champion should keep chasing the frontier or pivot to services — the exact ambivalence this report's own two-path framework was written to resolve, and hasn't yet.