Radical Numerics emerged from stealth on June 15 with $50 million in seed funding led by Emergence Capital, with participation from Obvious Ventures, Triatomic Capital, Factory, and First Spark Ventures. Patrick Collison joined as a pre-seed backer. The founding team — Eric Nguyen (CEO), Michael Poli (Chief AI Scientist), Stefano Massaroli (President), and Armin Thomas (CTO) — previously built Evo, the first model capable of reading and writing DNA at scale, and they describe the new lab's mission as building 'general biological intelligence': a single architecture that learns directly from DNA, RNA, proteins, and downstream biological data.
The bet is that the same pattern that worked for language — pretraining one large model on broad data, then adapting it to many tasks — can be repeated for biology if the data substrate is the right one. Today's drug-discovery AI tools are mostly narrow models trained on individual modalities: protein structure, molecular property prediction, ligand binding. Radical Numerics is arguing that the unifying layer is the genome and its information flow, and that an architecture trained at that level can generate novel functional sequences across the whole stack, rather than rerank candidates produced by traditional pipelines.
The launch fits a broader shift in AI-bio funding through 2026, where investors are moving past chatbot-style biology assistants and writing larger checks into infrastructure plays — Iambic, Generate Biomedicines, and now Radical Numerics. It also lands as the field debates whether the next biology breakthroughs will come from frontier labs adding bio capability to general-purpose models, or from focused labs like this one starting from biology and building inward. The seed round is on the small side for an AI lab, but the team's prior credibility on Evo gives them an unusual amount of technical optionality for the capital.
For learners interested in the AI-bio intersection, this is a useful reminder that the most consequential applications are not always the most visible ones. Designing protein binders or DNA regulatory elements is harder to demo than a chatbot, but it is closer to the kind of work that produces durable scientific and economic value. If you are choosing a specialization, the intersection of generative modeling and wet-lab biology has fewer practitioners than it deserves, and that imbalance is unlikely to last.