AI-Native Discovery Engines

Scientific discovery is still shockingly manual. A pharmaceutical company spends $2.6 billion and 10-15 years to bring a single drug to market, with a 90% failure rate in clinical trials. Materials scientists test maybe 100 compounds per year when millions of combinations exist. The bottleneck isn't ideas — it's the speed of the hypothesis-experiment-analysis loop.

Foundation models for chemistry and biology (AlphaFold, ESM-3, molecular transformers) reached practical accuracy in 2024-2025. Automated wet labs and robotic experimentation platforms can now execute thousands of experiments per week. The missing piece — an AI system that designs experiments, interprets results, and proposes the next hypothesis — is now technically feasible for the first time.

The drug discovery AI market alone is expected to exceed $10 billion by 2028. Recursion Pharmaceuticals (market cap $3B+) proved the model works but focused on biology. Materials science, agricultural chemistry, and industrial biotech are wide open. Companies like Insilico Medicine have drugs in Phase II clinical trials designed entirely by AI, validating the approach.

This is a deep-tech play requiring founders with PhDs or significant research experience in computational chemistry, biology, or materials science. The ideal team pairs a domain scientist who understands what experiments matter with an ML engineer who can build closed-loop optimization systems. Prior experience with lab automation or high-throughput screening is a strong signal.

Jon Xu sees this as one of the most consequential categories in the batch. The key insight is building 'closed-loop' systems — AI that doesn't just predict but actually runs the full discovery cycle autonomously. The winners will compress years of R&D into weeks, fundamentally changing how humanity discovers new drugs, materials, and chemicals.