Exclusive: The researchers who built AI-generated DNA just raised $50 million to reinvent biology

Radical Numerics is betting that the future of biodefense and AI drug discovery lies in understanding entire biological systems—not just individual molecules.

Eric Nguyen was a perpetual student, much to his parents’ chagrin. 

After receiving a master’s in engineering at Cornell and doing a stint training AI to interpret and understand the visual world, Nguyen enrolled in a Stanford bioengineering PhD program specifically to find a problem worth fighting for. 

“I basically went back to the PhD to look for purpose,” he told Fortune. “I wanted to find something that I thought I could contribute to, that if I didn’t work on it, nobody else would.” He found it in DNA.

Nguyen’s startup, Radical Numerics, emerged from stealth with a $50 million seed round led by Emergence Capital, Fortune has learned exclusively. Obvious Ventures, Triatomic Capital, Factory, and First Spark Ventures also participated. Patrick Collison, the CEO of Stripe and cofounder of the Arc Institute, backed the company at pre-seed.

Radical Numerics teaches AI to read, write, and reason in the language of biology—not just DNA, but RNA, proteins, and every other molecule that makes living systems work, all at once, in a single model.

The company’s founding team—Nguyen, Michael Poli (chief AI scientist), Stefano Massaroli (president), and Armin Thomas (chief technology officer)—are among the researchers who created the field of generative genomics. Three of the four previously built core technology at Liquid AI, an MIT spinout creating new AI model designs.

Together they built Evo and Evo 2, the first AI models capable of generating DNA sequences at scale, trained on the genomes of more than 100,000 species. Last September, researchers using Evo’s open-source weights produced the world’s first fully AI-designed functional virus (it was harmless to humans). That milestone is what pushed the team to build a company. 

“It still wasn’t being picked up in the way we thought it would,” Nguyen said of the academic work. “So we basically said: ‘We have to show the recipe.’”

The overall AI drug discovery market is projected to reach $25 billion by 2035, and competitor Ginkgo Bioworks recently signed a five-year AI platform deal with Google Cloud. 

But most AI biology companies today are single-modality, like Isomorphic Labs for proteins or Inceptive for RNA (which just signed a deal with Alnylam potentially worth $2 billion). Radical Numerics is instead betting that the bottleneck in drug development is about failing to understand an entire biological system.

“Getting the drug made won’t be the bottleneck forever,” Nguyen said. “You have to understand the whole system.”

The company has two early commercial partnerships: one applying its multimodal model to pancreatic- and multi-cancer detection, and one with a national laboratory to detect and characterize pathogens, including AI-generated ones. The revenue model is still taking shape but is a mix of API licensing, fine-tuned proprietary models for pharma partners, and milestone payments. 

“No one has figured out the right business model for how AI companies commercialize in life sciences,” Nguyen argued. “If anybody says they have a formula, they’re just full of it.”

There’s a catch baked into the entire enterprise. The same models that could accelerate cancer diagnostics could also lower the barrier to designing biological weapons, and Radical Numerics knows it better than anyone, because its own open-source work enabled that first AI-designed genome. “The defense side is sorely losing the race,” Nguyen said. 

The company brought on Andrew Weber, former U.S. assistant secretary of defense for nuclear, chemical, and biological defense programs, as an advisor, and is partnering with a national lab specifically to build AI-powered pathogen detection. Future model releases won’t automatically be open-source.

Ninety-eight percent of the human genome is still not understood. Nguyen is betting the same technology that could one day explain it could also protect against those who might exploit it. That’s either the best argument for building Radical Numerics—or the most urgent reason to hope it works.

This story was originally featured on Fortune.com