Ex-OpenAI Science Chief Kevin Weil Raises $215M for Warp Sciences
Kevin Weil left OpenAI to build an AI-powered chemistry lab where robots learn from humans to synthesize drug molecules.
Funding · Source: SiliconANGLE
What happened
Six months after leaving his role running science initiatives at OpenAI, Kevin Weil launched Warp Sciences. The drug discovery chemistry startup emerged today with 215 million dollars in new funding. Lightspeed Venture Partners led the massive round. Warp operates as a contract research organization in the United States. They supply biotech and pharmaceutical companies with the specific molecules needed for their drug programs. Their first service is lead optimization analog synthesis. This process produces variations of a promising molecule so customer scientists can compare them and narrow in on a final drug candidate.
The company is building a highly automated lab to execute these physical experiments. Experienced chemists currently guide the work. Over time, the lab will take on more automation. TechCrunch reportedly noted that robots will learn lab tasks directly from chemists wearing cameras and sensor gloves. The physical data generated from these automated experiments will be used to train Warp's artificial intelligence models.
Warp will operate strictly as a contractor. Chief Operating Officer DJ Patil, the former White House chief data scientist, stated the company is not developing its own drugs. The founding team includes Chief Scientific Officer Sol Reisberg from Terremoto Biosciences and Chief Technology Officer Mano Paluri from Meta Superintelligence Labs. Institutional Venture Partners, Lux Capital, and Menlo Ventures participated in the round. OpenAI Chief Financial Officer Sarah Friar and Coinbase Chief Executive Brian Armstrong invested as individuals.
Key facts
- $215 million — Funding raised by Warp Sciences in their launch round
- 74% — Share of surveyed biotech companies with Chinese contracts in 2024
- 6 months — Time since Kevin Weil left OpenAI to launch Warp Sciences
Why it matters
Artificial intelligence models for physical sciences are bottlenecked by a severe lack of real-world data. Software engineering and mathematics advanced rapidly because training data was abundant and digital. Chemistry and biology only advance when someone physically makes a compound and tests it. Warp is building a massive infrastructure to generate this proprietary physical data at scale. They are turning the slow, manual process of chemical synthesis into a high-throughput data engine for their models.
The geopolitical landscape of drug development is shifting rapidly. The BIOSECURE Act, signed into law in December, restricts federal contracts involving designated foreign biotechnology companies. Previously, 74 percent of surveyed biotech firms relied on Chinese contractors for preclinical and clinical services. Warp offers a domestic alternative to protect intellectual property. They aim to match the quality of overseas labs without charging the typical boutique price premium associated with US-based contract research firms.
For builders
Physical data is the ultimate competitive moat
Software AI relies on scraped internet text. Science AI requires physical labs to generate training data. Founders who can bridge physical automation with model training will command massive premiums from investors.
Geopolitics drives enterprise buying decisions
The BIOSECURE Act is forcing US biotech companies to abandon overseas chemistry partners. Startups that offer domestic alternatives at competitive prices have a captive market. Enterprise buyers will pay to protect their intellectual property and comply with federal laws.
Sell picks and shovels over pure plays
Warp is avoiding the binary risk of developing its own drugs. They are instead selling infrastructure and molecules to pharma companies. Builders should consider contract business models that generate immediate revenue and proprietary data simultaneously.
My take
AI is moving from generating text to manipulating atoms. Weil knows that the next massive leap in artificial intelligence requires physical data that you cannot just scrape from the web. Building a robotic lab to generate training data is an expensive but necessary brute-force strategy.
Original reporting: SiliconANGLE. This is my rewrite and opinion.