The valuation tells you everything: AI's Cambrian explosion just entered its "give me $750M at seed" phase.
The Summary
- Kevin Weil, OpenAI's former CPO who led their Prism science initiative, is raising $150M at a $750M valuation for an AI science startup focused on gathering scientific data for AI models
- This follows a pattern: senior OpenAI leaders (COO Brad Lightcap, CEO of applications Fidji Simo, Sora head Bill Peebles) are all heading for exits to build "something new"
- The shift from chatbots to science represents AI's next economic frontier, with ex-OpenAI and Google researchers launching autonomous experiment platforms and drug discovery tools
The Signal
Weil ran Prism at OpenAI, the workspace designed to put frontier models directly into scientists' hands. Now he's raising at three-quarters of a billion dollars before the product exists. That valuation isn't based on revenue or users. It's based on positional authority and a thesis about what comes after ChatGPT.
The thesis: AI models are data-starved for science. LLMs scraped the internet. Image models ate Getty and Shutterstock. But high-quality scientific data, experimental results, lab notes, the actual messy process of discovery, that's locked in university servers, pharma databases, and researchers' heads. Weil's pitch, according to sources, centers on gathering this data for AI models. Translation: he's building the scientific equivalent of what Common Crawl did for language models.
"AI's next frontier is shifting from chatbots to science."
This isn't abstract. Consider what's already moving:
- Google DeepMind's AlphaFold cracked protein structure prediction
- Recursion Pharmaceuticals is running 2.2 million experiments per week with AI-designed protocols
- Atomic AI raised $35M to design RNA-based medicines using foundation models trained on RNA structure data
The pattern: science moves slower than software because experiments are expensive and expertise is rare. AI agents can run simulations, design experiments, and interpret results faster than human researchers can write grant proposals. But they need training data that doesn't exist at scale yet.
Weil knows this because he watched it at OpenAI. Prism was supposed to be the interface layer, the ChatGPT for scientists. But interfaces don't matter if the underlying models haven't learned how science actually works. You need the raw material: failed experiments, marginal improvements, the 99% of research that never makes it into published papers.
The Implication
Watch where the senior OpenAI people go next. Lightcap, Simo, Weil, they're not retiring. They're placing bets on what the next $10 billion AI company looks like. Weil's bet: it's not another chatbot wrapper. It's infrastructure for scientific discovery at machine speed.
If you're in pharma, materials science, or any field where experimentation is the bottleneck, this is your signal. The AI agents are coming for your lab notebooks. The question is whether you're building the data moat or selling it.