Founded in 2022, ElevenLabs is an AI voice generation startup based in London. It competes with the likes of Speechmatics and Hume AI. Sopa Images | Lightrocket | Getty Images
Artificial intelligence companies are the hottest ticket items in today's startup ecosystem but the pace of change is dominated by developments at OpenAI and Anthropic. For startups building on top of their models, it's sink or swim. With the U.S. currently surging ahead in the large language model (LLM) race, which demands huge checks, Europe's opportunity lies in building tools that make AI useful, which is known as the application layer. "That's also where we think most of the profit will be made in the future," Robert Lacher, a founding partner of Visionaries Club, told CNBC's "Squawk Box Europe" earlier this year. Generative AI companies clinched $49.2 billion in venture capital (VC) investment in the first half of 2025, surpassing 2024's $44.2 billion across the whole year, according to consultancy EY. The U.S. is responsible for the majority of that, accounting for 97% of deal value and 62% of volume; Europe represented just 2% of value, but 23% of volume. Risk appetite among VC investors on the continent is typically lower than in the U.S., while market fragmentation has long caused challenges for startups looking to scale quickly. Hungover from the 2021 tech boom and amid an economic downturn, steady growth and sound business metrics have also come back into focus in Europe. AI is still drawing eyeballs but it pales in comparison to the U.S.
watch now
Now, frequent updates of AI models like OpenAI's ChatGPT and Anthropic's Claude are pushing companies built on top of them to iterate faster or risk falling behind. Europe does have its own LLM company – Mistral, the French startup that has raised 1.7 billion euros ($2 billion) in capital so far, including from Dutch chipmaker ASML – that is positioned as an open-source competitor to OpenAI, but there's still a lot of ground to cover. "The speed of innovation, speed of product velocity, speed of distribution, actually ends up winning over everything else," Bryan Kim, a partner at VC firm Andreessen Horowitz, told CNBC's "Squawk Box Europe" on Thursday from Italian Tech Week. Sweden's Lovable, a "vibe-coding" platform that enables others to build apps and websites with AI, and AI agent startup Sana are examples of such companies putting AI to use. Meanwhile, London's AI video generation startup Synthesia and synthetic audio company ElevenLabs, also have specific AI applications. The latter did, however, later build its own LLM.
watch now
But "what does it mean when the product and technology you're actually relying on changes every month. How do you move any slower than that and expect to win the game?" Kim said. "What I came around with is, actually, momentum is the moat at this current juncture of AI development. Maybe we'll get to a point where the model layer stabilizes it a little bit, and then we could talk about other things, but, right now, momentum is the only moat that I see," he added.
Building the next Spotify