Novartis Benchmarks Isomorphic and Chai

This shows that pharma buyers are treating AI drug discovery vendors less like sole strategic partners, and more like parallel test beds. Novartis is running Isomorphic on small molecule programs while also using Chai for antibody design, which lets Novartis compare model output, lab hit quality, and team workflows side by side, then keep the best ideas inside its own discovery engine.

• Isomorphic’s Novartis collaboration began in January 2024 on three undisclosed small molecule targets, and expanded in February 2025 by up to three more programs. That pattern suggests enough internal satisfaction to deepen the relationship, but not to make Isomorphic exclusive.
  1 isomorphiclabs 2 googleapis
• Chai is positioned differently. Its product is built for de novo protein and antibody design, and its own materials identify Novartis among adopters. In practice, Novartis can use Chai on biologics problems while running Isomorphic on overlapping discovery work, then compare speed, binders, and downstream wet lab success.
  3 chaidiscovery 4 sacra 5 chaidiscovery
• This is the same pressure coming from incumbents like Schrödinger. Bunsen layers AI over software that pharma teams already use, so large buyers can sample frontier models from startups without replacing existing workflows. That makes every vendor easier to benchmark, and harder to entrench.

The next step is that big pharma will keep a portfolio of AI discovery suppliers, while pulling more model evaluation, assay data, and prompt workflows in house. That shifts advantage toward vendors that either plug into daily scientist workflows, or produce repeated clinical wins that are hard for partners to copy.