Rivercell raises €22M seed to build an AI virtual cell

Rivercell, the Paris startup generating its own cell-response data to train an AI model of the human cell, has raised a €22M seed round led by HV Capital, with HCVC, Alven and Bpifrance Digital Venture participating.

The company came out of stealth on 7 October with one of the larger French seeds of the autumn: €22M, about $25M for the dollar crowd. HV Capital, the German firm, led the round; HCVC and Alven bring Paris deep-tech and software DNA, and Bpifrance Digital Venture puts the French state’s venture arm at the table from day one.

The money goes three ways, in the company’s own words: scaling its proprietary data generation platform, expanding its automated wet lab in Paris, and launching its AI Virtual Cell programme – a model meant to predict, in silico, how a human cell responds to a drug or a genetic change before anyone runs the physical experiment.

“A world model of the cell is the opportunity of the century in medicine,” said co-founder and CEO Yann Fleureau in the announcement.

From heart rhythms to the cell itself

Fleureau has shipped regulated AI before: he co-founded Cardiologs, the AI cardiac diagnostics company Philips acquired in 2021, and started Rivercell in summer 2025 (European Biotechnology, October 2026). Co-founder and chief scientific officer Eric Durand joined in 2026 with a CV that reads like a tour of European bio-AI: director of oncology data science at Novartis, chief data science officer at Owkin, co-founder of foundation-model company Bioptimus. The scientific advisory board pairs Fabian Theis of Helmholtz Munich with EPFL’s Charlotte Bunne.

The thesis is that the bottleneck in AI biology is not the model but the data. Rivercell runs its own automated lab to generate interventional, time-resolved, multimodal single-cell data – watching cells change over time and under treatment, rather than training on public snapshots – and aims the resulting model at oncology, immunology, rare disease and cardiometabolic conditions.

A $1.71bn market compounding at 30%

The market Rivercell sells into is small and steep: AI in drug discovery was worth about $1.71bn in 2024 and is forecast to reach $8.52bn by 2030, a 30.6% CAGR (Arizton, February 2026 update).

For scale against our own records, this was the third European bio-AI round above €20M in two days: WhiteLab Genomics raised a $26M Series B led by AVP for AI genomic medicines on 6 October, and Sensible Biotechnologies closed a $47M Series A led by Oxford Science Enterprises the same day. A seed within a million euros of a Paris Series B announced the day before tells you how early investors now price data-generation moats. Two of the three rounds belong to Paris companies, and our fundraising data logged all three before any of them had a week-old press page.

Paris is quietly stacking bio-AI infrastructure

So why does a seed this size land in Paris rather than Boston? Partly because the people do: Owkin, Bioptimus and Cardiologs built a bench of operators who have already trained models on biology and sold the results, and that bench is now recycling into infrastructure plays. I keep hearing that European capital will not pay for AI infrastructure before revenue; three rounds in two days argue otherwise. The loudest virtual-cell programme to date has been the Chan Zuckerberg Initiative’s in the United States – Rivercell is one of the first European teams to raise serious capital to build one around its own data engine.

The good news is that nothing in this round needed importing: the data comes from a Paris wet lab, the lead sits in Germany, and the state co-invests through Bpifrance Digital Venture rather than watching from the balcony. What we are seeing, round by round, is European bio-AI moving from borrowed models to owned data. What to watch next: whether the AI Virtual Cell programme publishes benchmarks, which pharma partner shows up first, and how fast the Paris lab hires.

Remember: the last French team Fleureau built ended up inside Philips. The ambition this time is to own the platform, not to sell the feature. Watch the labs – the models that matter next in medicine will be trained in one of them.