Why Europe Must Build Sovereign AI Or Risk Irrelevance.
Remember AI 2027?
The researchers behind that project are back with another great piece, expanding their visions about possible futures with AI developments, available at ai-2040.com. Instead of forecasting the future, they propose several alternative paths, indicating which ones are desired. This time around, the team proposes several parallel timelines with each of them changing at key decision points. I went through it. It made me think about geopolitics. I became increasingly aware of the role of Europe in the advertised scenario, dubbed Plan A. The thing is, Europe's role is close to non-existent. Let me explain why is this the case and what has to change about Europe's approach if people across the whole world are to benefit.
But first a brief recap of ai-2040.
The research outlines different plans, A, B, C, D and S, and their outcomes. The timelines diverge in 2029, when an important decision has to be made. The scenarios stretch all the way to, as you might have guessed, 2040. Plan A, i.e. the plan that the website champions as the best scenario is the Verified Slowdown. Plan B is slowing down and forcing China to do so as well (as U.S.), C is racing and using the advantage to develop safety measures, etc. Let's focus on Plan A, as that is the path most vigorously discussed, and it is proposed as the best solution. The timelines are presented mainly from the perspective of USA, which is fair given that the majority of AI development is currently concentrated in the USA, and the USA is, well, the USA.
ai-2040 talks about the existence of two workforces, working in parallel, one of them being humans and the other one being AI agents. It acknowledges that current AI Agents capabilities are still limited, but it predicts they will get better and increasingly become a more and more important part of the global economy. In particular, they will help the economies that have access to AI agents powerful enough to do any meaningful job. Under Plan A, U.S. and China become the two most important players in the AI race and they get to decide what the rest of the world has to agree to. They get to decide the future. In the wishful scenario they agree to a form of "Mutually Assured Compute Destruction" (MACD), in which they don't have to rely on mutual trust, but on performing checks instead. This agreement is largely based on the MAD strategy composed for deterring the use of weapons by retaining as much weapons to destroy the enemy, while the enemy holds onto the amount they need to destroy you. This approach acknowledged that complete destruction of nuclear weapons was not realistic, but there were paths to safety. AI safety research draws a lot of parallels with MAD.
What was striking for me in this scenario is the lack of virtually any role for Europe in it. It got me thinking...
Let's talk about Claude Mythos for a second.
This is an important piece of the puzzle. Anthropic announced Claude Mythos on April 7, 2026 claiming it to be a massive upgrade in existing model capabilities. At the same time, the company made it clear that this model will not be released to the general public due to the security risks it carries. They literally claimed their model carried too many cybersecurity and biology misuse risks for them to release it, and they decided to the access to it to chosen companies anyways. After the hype died down a little, experts from the industry expressed doubts whether the model was as capable as the company claimed, and some people said that the gated for safety release was a marketing stunt. Whatever it was, it portrays an important dynamic. When a newly built model is very capable, it might be released only to a number of hand-picked companies and they will be the ones to reap the benefits. Importantly, Anthropic disabled their most powerful Fable 5 and Mythos models after a U.S. government export-control directive ordered it to stop access for foreign nationals. This shows even AI companies are not above the law and governments can certainly make them bend.
Who gets to decide the future?
It increasingly seems like the access to the most capable AI models will be an important factor in shaping the economies of tomorrow, and the people responsible for granting or restricting this access will exercise a great power in deciding who gets to grow and who will have to shrink instead. The perspective presented by ai-2040 shows it clearly. The markets might shift to be an outcome of automated work performed by AI agents and robots, in which scenario having access to them matters a lot. Naturally, the power connected that comes with AI dominance concentrates in the U.S., as of now.
Hoog is a Dutch creator which recently presented an important European perspective that inspired me to think about the issue of AI access across the globe. In their recent video, Hoog presents a scenario that is bound to happen unless Europe does something remarkable. They present a scenario similar to the one of ai-2040, but give it a different framing. It is their perspective that prompted me to think about these questions seriously.
AI champions across the globe.
The landscape is changing rapidly but there are a few companies which managed to stay on top for the past few years in the capabilities race. Most notable companies working on AI in the U.S. are OpenAI with ChatGPT, Anthropic with Claude family of models, xAI with Grok, Google with Gemini, Meta with Llama family and plenty more companies whose models are less capable but could become influential in the future. They certainly have their champions, and that includes not only the companies, but also talent, data centers and the whole infrastructure around it. China is slowly closing in on the leaders of the race with their champions including DeepSeek, Zhipu ai (known by it's website Z.ai, the company behind GLM 5.2), Alibaba with Qwen, MiniMax with their models, Moonshoot AI behind Kimi assistant and many more examples. It suffices to say that these two countries have their say in the AI space.
Europe doesn't compare.
What's left in Europe is surprisingly little. Firstly, Mistral is based out of France and remains the most capable European foundational model, that is besides DeepMind's work whose interests after the acquisition by Google I consider American. OpenAI has thus far raised about $180B over it's funding rounds, and Mistral has pocketed $3.05B, roughly 60 times less, while being the biggest model out of the continent. Comparing funding is not the best metric for model's capabilities, but if you look at the sheer number of AI companies coming out of the U.S. and China as well as the capability benchmarks like Code Arena the situation is quite obvious (Mistral's highest ranked model is 80th at the time of the writing of this post, with U.S. and China ahead). Europe has a lot of catching up to do. In the next paragraphs I will explain why this is bad thing not only for the people from Europe but for all of us, what levers can Europeans pull to change their course and what could happen as a result.
Why would anyone outside of Europe, care about Europe?
We live in a world which is very interconnected, and goods and services can flow through continents' boundaries. Importantly, international cooperation happens because of legislations and governance is in place to enable them. Europe has been a leader that adopted multiple innovative rules which propagated through the legislation systems of the world. It put liberal democracy and parliamentary institutions as a benchmark, while actively supporting election observation and democracy-building abroad. The continent's European Union showed that former rivals can be reunited by cooperating between their markets and institutions, creating peace as a result. European privacy and platform accountability rules have had impact on numerous companies and regulators across the world, because internationals tend to adopt the strictest standard across markets, which happened to be Europe. There was also a lot of climate and environmental initiatives that drove climate diplomacy and standards across the world, helping enforce rules on emissions, sustainability and green industry.
Europe can bounce back.
The most important asset the old continent currently has is its largest company by market cap, the Dutch extreme ultraviolet (EUV) lithography machines producer, ASML. It is the only company capable of producing machines necessary for modern semiconductor industry, and by choosing who the company sells their machines to, has a huge influence on the global markets. Importantly, the key talent is also within Europe with top-tier technical universities like Imperial College London, ETH Zurich, Technical University of Munich among the leaders, it has the academic institutions to attract the best researchers in the world. With necessary revisions of how research is done at scale in those universities (particularly by spreading the model of universities like ETH or EPFL to allow professors to supervise more student efficiently) and appropriate grants coming from the governments and EU, it could catch up by a lot. Innovative environments which were created among the leading universities, with the most notable startup hubs in London, Paris and Munich could rival their oversees rivals, if perhaps more money was made available to them. The money needed for financing innovation is in Europe, but being held in pensions funds, it is discouraged from being invested in risky ventures. Proper regulations would need to be introduced to treat these problems.
Criticizing is always easier than giving helpful feedback.
People that have the power to change things often get stuck in negative loops that make them ramble about how stuff doesn't work and how they wish somebody finally fixed it, never realizing that they are in power to do exactly that. Capabilities of AI models increase, more and more could be gatekept. With these changes, there might finally be a rally around the flag effect in Europe which will make its residents stand up, take a stance and get to work instead of arguing. Let's hope we don't have to wait for dramatic loses before a breakthrough comes around, making people gather and cooperate.