"Europe is out of the AI race." It is the most repeated and least accurate narrative of the moment. We genuinely do not have an OpenAI, a Google DeepMind, or an Anthropic. And yes, the American tech giants raised more than 23 times as much investment capital in 2025 as China. The gap with Europe does not make for cheerful reading.
Anyone looking purely at those numbers concludes that we are watching from the sidelines while the rest of the world builds the future. That story is true for only half of the AI race. And as far as we are concerned, it is precisely the wrong half.
The development of AI is about two things:
- The Frontier race: who builds the smartest model? This is where the battle over strategic power and infrastructure is fought. On this axis, Europe is indeed lagging well behind.
- The Adoption race: who applies AI the fastest and the smartest in practice? This is the race that determines the actual productivity gains. And this contest is still wide open.
By looking only at the Frontier race, we make a historic mistake. With fundamental technological shifts — whether electricity or the computer — most of the economic value is added by whoever applies the technology. Not by whoever invents it.
The digital sector itself (software, telecoms, cloud services) accounts for only around 10% of GDP in developed countries. The real value of AI emerges when it lands as a tool in the other 90%: manufacturing, logistics, retail, and healthcare. There it is not called 'AI', but a more efficient factory, a more profitable shop, and a smoother supply chain. Look at that playing field and Europe holds an enormous trump card. European manufacturing accounts for 14% of our GDP, against not even 11% in the US. European industry has some 2.2 million companies and 30 million workers. That is where AI can add the most value. But let us be honest: factories where you can apply AI are not enough on their own. The gain is determined by the speed of adoption.
China gets it
Look at China and you see how the adoption race is played. The national "AI+" plan is not primarily chasing the smartest frontier model; it pumps the technology straight into industry and logistics. They do this not with bulky, billion-dollar super-models, but with vast quantities of small, specific, and cheap AI models. From Alibaba's open Qwen models alone, 100,000 variants have now been built.
For 90% of business tasks — reading an order, routing a customer query, quality control on the production line — a small, specifically trained model is perfect. It is considerably cheaper, much faster, and, not unimportantly, it runs on your own infrastructure. A smaller open model that you run locally (on-premise) and fine-tune on your own processes is in fact the ultimate asset. You do not lose it when a tech billionaire raises the prices or pulls the plug. It also makes complying with the EU AI Act and other regulation a great deal easier: your data stays within your own walls and the model is explainable.
There is no last mover advantage
That is the heart of it. So we need to get serious about adoption at the macro, meso, and micro level. The company — or sector, region, country — that starts applying AI today builds a lead that later entrants cannot simply buy back. Not with a better model, and not with a bag of money. By doing, you build up process data, your people learn to work with AI, and small successful use cases stack up. This learning process takes time, and time to wait for the "perfect moment" or "better models" is exactly what we do not have. The real price of slowness is not the technology you miss out on, but your competitor's learning capacity growing quarter after quarter.
In short: Europe does not have to win the frontier race to extract enormous value from AI. We are not out of the game yet. The clock is ticking, though, and the only variable you control today is how quickly you begin.
Our lesson: begin. Preferably small, but broadly supported. Prove the value of AI in the practice of your organisation. Scale what works.