Premise: “AI Maximalism” here is my shorthand for the diffuse view that frontier LLM models (where Europe is nowhere relative to the US labs) are key to our economic future, and our entire fate depends on whether we can adopt them superfast – because building them is not on the cards. There are nuances of course, and I like and learn a lot from the authors I mention in this piece, subscribe to their Substacks and have great respect for them. [i] So spare the vitriol. But I also think there are misunderstandings to be clarified from the perspective of the tech sovereignty movement, in which I am very involved (EuroStack Industry Initiative). Conversely am not well disposed towards the Europe 2031 dystopian novelette doing the rounds, which I believe to be ill-considered and harmful to Europe, “Igor” to hyperscalers’ Frankenstein (h/t Jay Latta for this metaphor).
The exploding debate around Europe’s tech dependencies in the wake of Anthropic’s decision to pull its advanced Mythos/Fable 5 models globally (won’t go over the circumstances, well known at this point) has been inevitably resonating with “wake up call”, “kill switch”, “must build our own”, and “sovereignty ahoy”. One major strand of this discussion has been the use of the episode (in itself of course significant, but not about Europe in the least) as the jumping off point for exhortations to Europe to “be realistic” in AI; accept we are “over” in terms of any hope to catch up with the US top AI frontier LLM labs; accept we are just a “middle power” and all sovereignty efforts are therefore just “theatre”; accept the best we can hope for is to be a “smart follower” (i.e. adopt current US frontier models at breakneck speed, as that is our only chance to create value with industrial applications); EVEN, given how short we are in compute capacity for these applications, set out multiple inducements to hyperscalers to build more datacentres in Europe in exchange for some commitment to supply Europe with compute that is becoming ever more scarce.
I will not discuss here the view that frontier LLM models are the only future, and thus Europe is doomed as an AI power (“no Fable class model” but “at most ‘only’ at the level of Opus 4.8 today” even if we started right now and gave it all we got for three years). I talk about other paths and possible architectures in my forthcoming podcast with Andreas Liebl of the appliedAI Institute and Foundation, out later this week (see forthcoming Escape Forward).
What I discuss here is a misunderstanding that is common to the position of “maximalist” believers in the supremacy of LLM AI frontier models about what the European tech sovereignty movement stands for. The misunderstanding finds expression in multiple ways:
- A dismissal of “Linkedin types” (presumably people like me) who advocate for European sovereignty but have not set out a clear plan for Europe to develop frontier AI models that compare to the top US labs (Open AI, Deep Mind, Anthropic);
- An assumption that the discussion about European sovereignty must be - by definition – about frontier AI, because frontier AI is the future and ultimately all that matters. The sovereignty movement thus must be about AI sovereignty, and as such it is a hopelessly ill-conceived and delusional wish to go to the mat with the US on frontier models (typical argument: you don’t have the billions and trillions required to generate a real “champion”, look even Musk failed, Meta is failing even throwing billions at it, and indeed spending “more than Germany spends on defence”, even if you did it would not generate returns, etc.);
- An assumption that what the “sovereignty types” are pursuing is “divorcing from the US ecosystem” (!) and therefore contempt for all our efforts as just “sovereignty theatre”;
- A declaration that “sovereignty is a valuable goal, but the clearest way to be sovereign is to be prosperous and rich”;
- A view that “sovereignty requires innovation first” (statement by the CEO of ASML at VivaTech, he is not an AI Maximalist but I mention it here as the “we need innovation!” mantra has been one of the worst platitudes and albatross around the neck of the tech discussion in Europe for years).
So let’s make a few things clear.
1. The European sovereignty movement is not focused specifically on AI or frontier AI. We care about Europe’s tech assets and capabilities along the tech stack: from chips to cloud to software to connectivity. AI is a VERY important user case (cloud + software + data). But it is not what we uniquely focus on. We want to see each of the pieces of the stack being invested in and growing. In Europe and for Europe (and of course for the world if we can sell globally). Because we DO have great businesses at various level of the stack, they are hemmed in and under-demanded and under-invested (we know all that, it’s the whole point) but they are technically capable and innovative. We want these hundreds of European businesses to flourish, not to shrink or be inevitably acquired. Across the Continent there is a huge amount of talent, Europeans are super smart and educated and inventive. But demand is constrained by fragmentation and hyperscaler capture, capital that we have in huge quantities is massively conservative. The purpose of the movement is driving demand and capital to these companies, which are not candidate frontier AI players, but are all components of our tech ecosystem. We are not a desert of talent and effort and capital. We have amazing stuff.
2. The ultimate purpose of the sovereignty movement is to assist the “Draghi mission” of growth and productivity improvement in Europe. Not “our values, our democracy”. It is also a reaction to the stubborn fixation of vociferous civil society and academics for “enforcing antitrust and regulation” which has done/is doing perfectly nothing for Europe for a generation. Draghi is clear that Europe’s laggard position in tech adoption and diffusion is a major cause of our productivity gap with the US. We have witnessed a significant debate among economists on whether the gap is real (and will be hosted by DG Grow with Luis Garicano and Sander Tordoir next week to debate it) – but the upshot (as summarised by the Economists and agreed with by Garicano) is “First, Europe is growing slower than America. Second, that is a problem: even if Europe can free-ride on America’s dynamism, as Team Krugman argues, that is no way to run an economy. Third, the AI era may be less forgiving than the last wave of tech-driven growth.” Agreed. AI is SUPER important. And its adoption will be critical for Europe’s productivity. Our main point though is that unless we own more of our factors of production (namely tech assets), and capture more value along our supply chain, we are syphoning off that value to non-European player. This matters because that revenue/profit is reinvested in US innovation and products, not by Europeans in Europe. We would like to retain a larger share of value added here please.
3. This objective is not fulfilled solely by “boosting our startups”, nor just focusing on “agentic AI startups” which are all the rage for our AI future. All of this matters, a lot. But we DON’T just have startups to support in Europe. We have tonnes of them, and it is absolutely true they need support and funding. BUT we also have hundreds, thousands of creative honest-to-goodness tech businesses from chips to hardware to software to everything in between that are there, have great products, and should sell more. They should scale up. Yes they need to integrate a lot better! But they are there. This is not instead of pushing the plethora of agentic AI startups which are crowding up the space. It is AS WELL.
4. So when the “AI maximalists” denigrate the European sovereignty movement, it would be good if they understood no one here is making the case we stand a chance to go to the mat with the US on AI frontier models. Is this fatal to us? As mentioned I discuss with Andreas Leibl other potential visions for Europe (like it or not). But as far as EuroStack and the tech sovereignty movement go, we are interested in a different issue, which matters for our economy as well: the health and prosperity of our entire digital sector. AI agents will transform the way business processes work, and this transformation is underway. But none of this makes European chips, cloud, software, redundant. We will have many different user cases where the demand for inference and compute and teams of agents is very high; many other European businesses and user cases are in a different place. And do not need the same configuration.
What else do I disagree with in the framing of the debate by “AI maximalists”? As mentioned much of the writing is thoughtful and does consider options, albeit again much from the perspective that the current LLM frontier models are inescapable and the way forward, and the sooner we accept that the better. However:
a. Luis Garicano in Silicon Continent advocates for a “smart second mover” strategy in which Europe accepts it is a Middle Power and adopts existing US models at speed to capitalise on opportunities for industrial and physical AI. As an economist who worked for years on rent extraction by US digital platform monopolies, I am worried the current vision that “the models are competitive, they will not extract value, the rents will flow instead downstream to hardware and upstream to implementers” (as discussed in my podcast with Luis and in his Silicon Continent piece on value capture) will not be the last word – rent extraction will manifest sooner or later. Then what.
b. As discussed with Andreas Liebl, frontier LLM models may not be the only path, and there are efforts underway not to throw in the towel – see what SPRIN-D’s “New Frontier AI” initiative is doing, pushing public and private capital and AI talent in Europe together. The Garicanos also do consider a scenario in which Europe puts a lot of effort to develop models “a few steps behind the frontier” and recognise that’s also going to be useful (a competitive fringe to LLM frontier models is important, and not all user cases require frontier and masses of inference).
c. There is also discussion of a more “moonshot” path in which Europe pools resources and finally does a “CERN for AI”, leveraging our strength in open source as a competitive advantage. Anton Leicht, who has written extensively on our vulnerabilities, also just put out a “Moonshot” paper where he goes systematically through what Europe and other middle powers would need to do / invest to stand a chance to develop a frontier model: educated guess, invest 500bn and counting. Appreciate the systematic assessment.
d. Least acceptable to me is the suggestion by Anton Leicht and indeed Luis/Pieter Garicano that “middle powers” like us should give incentives to hyperscalers to build more data centres in Europe, perhaps in the form of cheap/clean French nuclear energy, in exchange for promises they will guarantee to us the compute which is currently so scarce (“US allies should offer American hyperscalers favourable terms for datacentre buildouts – like subsidised energy prices or even outright energy access…in return for guarantees to always provide frontier capabilities”). Leicht recognises this may sound “submissive and uncomfortable”, but it is “necessary” and “reasonable” as we need frontier access right now. That of course depends on a view it’s “AI frontier or bust”. Even so, how enforceable would be all this? And why give it to them? Because they have the cash? The constraint is actually not funding when you speak to European datacentre builders. It is ONLY DEMAND: “will I have demand for our capacity, or will I just have Microsoft as a tenant”? Building datacentres without plugin demand first is suicidal. Giving our clean energy (e.g. France and its nuclear energy surplus) to hyperscalers locks up our energy, doesn’t create long lasting jobs, and reinforces hyperscalers’ entrenchment. Why on earth should we do that. We will build more ourselves if demand materialises. The constraint isn’t capital: it’s demand.
Overall: does this mean that the pursuit of “sovereignty” as described above is redundant and silly? I don’t think so. Again, we don’t just focus on AI. Driving demand and funding to European assets along the stack is important, the hundred companies labouring in Europe deserve to thrive, not just survive.
Finally, a word on why I am so dismayed by the Europe 2031 “novelette” that has been doing the rounds. The authors declare their intention was to tell the story of how Europe will be likely to decline into inescapable insignificance in AI and its economic future be hollowed out by 2031if nothing happens (i.e. business as usual). And as such, their motivation was to spur Europe into action. My real problem with it is not just the cringe tone and the predictability of the narrative (the motivated young French staffer in DG Trade’s digital unit, frustrated at the inertia and haplessness of her superiors; the German young decamper to Silicon Valley who constantly beats up on Europe as seen from there by his buddies, and ends up super rich when he sells his startup over there; Europe’s repeated failure to take any action that takes us to final doom). All of this is bad enough.
But what’s seriously worse is that (a) it’s word by word Silicon Valley cool aid, on stilts. Everyone who gives it a cursory glance in Silicon Valley loves it. Hyperscalers could have written it. The Silicon Valley superiority about Europe’s tech (“they just suck”) is indeed a major pillar of the piece – I am familiar with this rhetoric which is pervasive as it is (frankly) ignorant of much of Europe’s reality. I discussed this with (let’s say) one of the stealth contributors to the piece, and the discussion went along these lines: me - “how much do you know of the breadth and depth of Europe’s tech companies, we have 500 in EuroStack”; response “my Silicon Valley friends think European tech sucks, do you think you are smarter and know better than my Silicon Valley friends?”. Okay.
And what is the recommendation? Again several, but prominently “we are so behind we need to give hyperscalers’ incentives to build more data centres in Europe so they can guarantee us compute”. Why not Europeans? And how enforceable are these promises/contracts? “Oh if we become a significant portion of their demand they will never want to let us down and walk away”. Really. How nice. Depends on what Executive Order they face.
This is disturbing because it is EXACTLY what hyperscalers have been arguing around Europe. “We will build datacentres for you, dear customers, will never let you down”. Who captures the value? Who sets the direction of innovation? How credible is the promise? As Jay Latta fittingly says, all these ostensibly “pro-Europe” opuses may well end up being Igors bearing the tools for Frankenstein. Beware. What a coincidence this came out just ahead of the G7, “access to compute is what they are discussing today in Evian” – so cool.
In summary: the “AI maximalist” contempt for European sovereignty comes down to: “frontier AI LLM models are existential, and if you European “sovereignty types” are arguing for investing in your own models, that’s a total waste of time given you will never catch up. Accept you are at best going to be implementers and need to do everything it takes to secure compute”.
We are of course super interested in Europe’s AI future but do not advocate for “sovereignty in frontier LLMs” specifically – even less “decoupling” of any sort. This is a very confused point. Our vision of “sovereignty” involves building more of our own assets and capturing value along the tech supply chain, so that we do not just skim rents off the top, as one way to help growth and productivity and prosperity. We see AI as a critical user case – hardware, cloud, software, data. We absolutely must build and power up these assets in Europe. It is not wasteful substitution. It is the infrastructure backbone of our digital economy. Which we cannot leave entirely in non-European hands or our own asset base and skills will atrophy further into a digital colony.
No one here is trying to build another Microsoft or another AWS or Open AI or Anthropic - that’s a straw man. But we need to build and throw everything at it. And by the way, Europe is NOT a Middle Power. That framing is super unhelpful to the mindset. Europe is a SUPER power in terms of ECONOMIC muscle and capabilities. There is value and valour in pushing European demand and capital to support the side.
(A final postscript: I quoted ASML’s CEO at VivaTech saying “to get sovereignty you need innovation first, we need to make sure innovation can happen in Europe”. The rest of the interview is actually surprisingly on message with EuroStack as it emphasises that we need above all DEMAND. But the initial foray into “innovation first” is classic European enterprise CEO-line. All captains of industry say that. We have more “innovation programmes” in Europe that you can throw a stick at. I can name dozens. But it is not the point. WE DO have tonnes of innovation in Europe. We don’t turn that into products and markets because demand is captured by hyperscalers and funding (which we have IN SPADES) therefore is awol. This is the only issue here. Sovereignty is about reversing that. No one will scale if demand is inert and not going there. This is what the sovereignty movement is trying to activate).
[i] The main pieces I refer to here are the Silicon Continent substack by Pieter and Luis Garicano (in particular, The smart second mover - by Luis Garicano, The Smart Second Mover, Part II - by Luis Garicano, Three theses on AI value capture - by Luis Garicano, and Nineteen thoughts on AI and Europe). And the Threading the Needle substack by Anton Leicht, in particular Cut Off - by Anton Leicht - Threading the Needle, The Moonshot - by Anton Leicht - Threading the Needle and Anton Leicht on X ).