A firm with four workers and a functional AI product is choosing where to incorporate in a serviced office in King’s Cross, just a short stroll from Google DeepMind’s London team. Dutch and German people founded the company. The product was manufactured in Berlin. They’re going to raise a Series A round, the product works, and the early customers are satisfied. They are attempting to determine whether to proceed with that round as a German GmbH or a UK limited company; in contrast to two years ago, the EU AI Act is one of the factors in that decision.
The most extensive attempt by any country to govern artificial intelligence as a separate legislative category is the EU AI Act, which went into effect in August 2024 and is currently implementing its compliance timeframes across various risk categories. In theory, the risk-based paradigm—which categorizes AI systems from minimal to intolerable risk, with compliance requirements calibrated to category—is a reasonable framework for a technology that is concurrently employed in high-stakes and low-stakes applications. In actuality, the compliance burden is significant, costly, and front-loaded for a tiny startup developing a product that may fall into one of the higher-risk categories—exactly the stage of a business’s existence when funds are few and time is of the essence.
The conversations are driven by the figures that compliance attorneys are presenting to early-stage founders. Before any product revenue exists to offset it, the legal overhead for a startup deploying an AI system in a regulated sector can range from €50,000 to well over €200,000 annually due to risk assessments, technical documentation, conformity assessments, and human monitoring needs. The EU AI Office has indicated that it plans to deploy its enforcement powers, which include the ability to demand product recalls and impose fines of up to 7% of global annual sales. The exposure is not hypothetical for a Series A firm with €2 million in revenue and a product that meets the Act’s high-risk criteria.
London’s proposal to founders in this role is a combination of regulatory and non-regulatory. The political consensus following Brexit resolved on framing the UK as a pro-innovation regulatory environment, at least in technology, which is why the UK government’s approach to AI regulation is purposefully different from Brussels’. This is not because the UK is unaware of the risks associated with AI. The UK does not have an AI Act.
Rather, the current sectoral regulators—Ofcom for media-related products, the Financial Conduct Authority for financial applications, and the Information Commissioner’s Office for data-related inquiries—apply their current frameworks to AI use in their respective fields. As a result, early-stage products have greater flexibility, there is less paperwork to complete before launch, and it is possible to make changes in the market rather than in compliance meetings.
Since Brexit, the infrastructure of venture capital that makes London practically appealing for such a transfer has grown significantly. The UK attracted over £10 billion in technology investment in 2023, and the concentration of AI-specific funds in London has increased as the sector has become a priority for both domestic and foreign investors, defying the prediction that London’s financial and technology sectors would hollow out after leaving the EU single market. Moving from Berlin or Amsterdam to London gives a business access to a venture ecosystem that is less concerned with the compliance-first concerns that investors in the EU market now have to deal with and more at ease with earlier-stage AI bets.
The issue of market access, which London is unable to fully address, is the practical challenge. Under the terms of the EU AI Act, a UK-incorporated AI firm does not immediately have the right to sell within the EU single market. Businesses with significant EU aspirations usually deal with this by creating a subsidiary structure, such as a UK headquarters or an EU operational entity, which adds complexity and expense. Based on their particular product, target market, and growth trajectory, entrepreneurs are determining whether that complexity is less than the option of being incorporated in Germany and managing the full EU AI Act compliance burden. Not everyone has the same response.
