Inside ai-PULSE 2026: What’s holding Europe back from turning AI research into business success?

ai-PULSE came back to Station F in Paris on October 13, with more than 2,000 people turning up for Scaleway’s annual event. The agenda was built around three themes — ENTER, SHIFT and CTRL — and the focus was less on shiny model tricks than on a stubborn European problem: the research is real, but the money-making part is much harder.

That tension came through again and again. Sébastien Le Roy of Serena said AI agents are getting good at narrow tasks, but handing them full business processes is another matter entirely. A system that works 80% to 90% of the time can look great in a demo, he said. In production, that same error rate can be a disaster when it repeats across thousands of important decisions. Enterprises want software that knows when to act on its own, when to ask for help, and how to recover when it goes wrong.

The same basic problem shows up in embodied AI. Andreas Geiger, who leads KE:SAI and is a professor at the University of Tübingen on leave, said demos are easy, but that no self-driving taxi company has yet turned profitable because the costs of research and engineering still outrun the revenue. Le Roy drew a rough line between sectors: voice is already a business, robotics is getting there, world models are a bet on the next computing paradigm, and quantum lives on a different clock entirely. Voice is also the clearest proof that customers will pay; he pointed to ElevenLabs, which reached a $22 billion valuation earlier this month.

Scaleway’s Damien Lucas made the other half of the case: sovereignty matters, but it does not close deals by itself. His pitch is sharply European — no subsidiary, employees, or shareholder outside the EU — yet he said customers will only switch if the features match what the major US cloud providers already offer. Price helps, too. Scaleway does not charge egress fees, which Lucas said can make up 20% to 25% of an AWS bill, and he said LVMH and Airbus expect to save 20% to 30% on migrated applications.

The funding side looks healthier than it did a year ago. Anaïs Monlong of IRIS said seed and Series A money is flowing again after a long slowdown, and that startups which spent 2024 and 2025 rebuilding around generative AI are finally coming to market. There are bigger rounds returning too, including Paris-based Gradium’s $100 million seed in July and IRIS’s co-led €35 million round for Implicity in September. But open research still leans on donations more than venture cash, and Geiger was blunt about that. VC is easy to find. Donors for open scientific work are not.

That is the European story in miniature: the science is there, the talent is there, the customers are there, and still the continent keeps tripping over the same old handoff from lab to revenue.