Fredrik Erixon / Apr 2026
Image: Shutterstock
At an event that we hosted a few weeks ago on European competitiveness, I mentioned in passing that ever since the AI Act was decided in the Spring of 2024 I have kept a tally of European AI companies that have left for other regions because of Europe’s AI overregulation. My tally in March was 74.
An enterprising member of the audience from the German tech community sprang into action and surveyed his European start-up network. Was my number right? He got back to me last week with a new one: he had 136 examples of start-ups that had left Europe in the space of the last 18 months because of the AI Acts remorseless complexity.
Unlike many European tech entrepreneurs who move their business registration elsewhere but keep the actual development in Europe, these are people who have relocated it all – lock, stock and barrel. This is quite a feat for a regulation that had not been fully implemented before the Act was paused for omnibus reform!
And when I read the Bloomberg interview this week with Siemens boss, Roland Busch, I was not convinced that all the declarations of a less burdensome AI Act will stop the tech emigration. “Much of the company’s €1 billion investment in industrial AI will be redirected to the US due to Europe’s regulatory burden”, Buch said. He doesn’t think the new AI rules will make life much simpler for his lot – developers of industrial AI.
In his last pamphlet, the 19th century French economist Frédéric Bastiat captured the epistemic problem in Europe’s regulatory bonanza of digital technologies: “What is Seen and What is Not Seen”. There are many things we have chosen not to see. Can you remember reading anything in the EU’s analyses and impact assessments that start-ups were going to leave because of the AI Act? I don’t. And I know why. The people running the show back then weren’t interested in the possible effect.
Lately, I have had similar problems with the Digital Markets Act (DMA), the landmark regulation from 2022 that was billed to “reorganise the digital space”. Well, of course it didn’t. It was implausible to think it ever could. But has it achieved outcomes that have had meaningful impacts on competition and consumers?
The whole DMA is now under its first mandated review, and I had hoped that EU regulators would focus on the impacts. Very few seem to do. We know some aspects – parts that are chose to be seen, to use Bastiat’s metaphor. The Commission has published a few reports on implementation and compliance, and platform companies have reported their own actions. But what about the “not-seen” aspects?
In a new Policy Brief, colleagues and I have pointed to some issues that are important but routinely unobserved. One can chose any type of outcome or effects-based metric, but the general impression is that the DMA mostly is a nothingburger – a seemingly great event but one that has not changed much. It has forced high compliance costs on platforms. Some user experiences have improved; others have been damaged. Interoperability demands have bettered business access to platforms, but the effect is minimal. Most users stick to their old habits and choices of how they interact with different platform services because they work well.
DMA enforcement also seems to be changing. As Dyuti Pandya and I point out in our Insight, the EU enforcers are increasingly taking new enforcement right into the core architecture of platforms and products. Enforcement is now less about behavioural conduct rules and more about product design and even price mandates – like in its recent specification proceedings. Is DMA enforcement effectively making the regulation into a product and price regulation?
In another Insight, Andrea Dugo and I went through some of the problems that arise when DMA enforcement seeks to micromanage product architecture. The DMA seems to have been administratively coronated as king or queen among EU regulations that impact on digital services. If mandated DMA enforcement bump into, say, cybersecurity or data privacy rules, it’s then the DMA that has right of way. As far as I can recall, however, this is not what legislators intended when they adopted the DMA.
It is again this old “Bastiat” problem – the consequences that are seen and those that are not seen because we chose to not look at them. Enforcement is now premised on the idea that effects and outcomes are immaterial, and it is rare that the European Commission motivates its actions with such evidence.
A different approach, colleagues suggested in another Insight, is that Europe now follows Japan and its new “DMA regulation”, and rather develops an effects-based enforcement style. This is also what our Policy Brief recommended: the DMA does not necessarily need reforming, but the enforcement needs an overhaul to avoid enforcement drift, improve evidentiary standards, integrate with cybersecurity and privacy rules, and develop a lot more pragmatism on interoperability policies.
If the EU seeks to be less programmatic and more outcomes focused, the DMA could achieve better results.
