Bessemer outlines AI drug-stack layers

Bessemer Venture Partners published an April 12 note that breaks AI drug discovery into distinct commercialization layers rather than treating the category as a single market. The firm said the stack runs from biology-native data generation to model and software workflows to lab automation, and argued that competitive advantage will come from how companies connect those pieces. (bvp.com) Andrew Hedin, Marla Jalbut and Grace Dai are listed as contributors to the note, titled “Building biology-native data infrastructure for the AI era.” Bessemer wrote that falling compute costs and maturing models are shifting the bottleneck in drug development toward data infrastructure and experimental execution. (bvp.com) The note says “biology-native data, agentic workflows, and lab automation” will define the next generation of biotech companies. Drug development still takes more than five years from target identification to a clinical candidate in many cases, according to the Bessemer note, and nearly 90% of drugs entering clinical trials fail. The firm also said R&D costs per approved therapy continue to double every nine years, framing the economics that AI-focused drug companies are trying to change. (bvp.com) Between 2012 and 2022, about 200 AI drug-discovery companies raised a collective $18 billion, Bessemer wrote. That earlier wave produced a market centered on model builders and platform companies, but the new note puts equal weight on the upstream creation of proprietary biological data and the downstream automation needed to test hypotheses at scale. (bvp.com) Insilico Medicine’s June 2025 Phase IIa results for rentosertib are one of the examples Bessemer used to show clinical progress from AI-led discovery. The firm said the program was the first to generate clinical proof of concept where both the target and the molecule were designed using generative AI, and added that the company nominated a preclinical candidate after screening 78 molecules in 18 months at less than 10% of the average cost per approved drug. (bvp.com) In early 2026, GSK and Eli Lilly signed model-access deals that Bessemer cited as evidence of commercial demand for specialized AI systems. The note said GSK committed $50 million upfront to NOETIK for access to oncology models, while Lilly agreed to a mid-eight-figure annual access fee to Chai Discovery for biologics design. (bvp.com) insitro and Isomorphic Labs also appear in the note as examples of different positions in the stack. Bessemer said insitro’s combination of large-scale human cell data generation and machine learning had led Bristol Myers Squibb to nominate two additional ALS targets from their collaboration, while Isomorphic Labs has pursued partnerships with Lilly, Novartis and Johnson & Johnson worth more than $3 billion in potential value as it advances its own oncology pipeline toward first-in-human trials. (bvp.com) Bessemer has been making the same broader case across other 2026 healthcare and life-sciences writing. In a March 11 roadmap on life-sciences AI, the firm said pharmaceutical companies spend more than $150 billion across service providers and software and argued that AI-native challengers are emerging in drug discovery, trials, regulatory work and manufacturing. (bvp.com) On Bessemer’s site, the April 12 drug-discovery note remains the firm’s most specific outline of how data generation, software and automation fit together inside the AI biotech stack. (bvp.com)