Shift to closed loop AI discovery

The edge in AI drug discovery is moving from having a smarter model to owning a faster learning machine that can generate data, test molecules, and carry winners into the clinic. Xaira was built around that full loop, with nearly $1 billion at launch, a focus on protein and antibody therapeutics, and internal efforts like X-Cell and X-Atlas that widen it from molecule design into target discovery and perturbation biology.

• Recursion is the clearest proof that the market now rewards full stack platforms. Its November 20, 2024 combination with Exscientia joined large scale phenomics data, chemistry, and clinical programs into one company, showing that model layers are being bundled with wet lab and development capabilities.
  5 recursion 6 sacra 7 sacra
• Isomorphic Labs shows the next step in the same direction. It started from frontier structure models, then built major pharma collaborations with Eli Lilly, Novartis, and Johnson & Johnson, which means the model is no longer sold as software alone, it is tied to real discovery workflows and downstream drug programs.
  8 isomorphiclabs 9 isomorphiclabs
• This changes what counts as defensible. Open protein models and fast copycat interfaces can spread quickly, but petabyte scale cellular datasets, lab automation, and owned therapeutic programs are slower to replicate because they require years of experiments, capital, and clinical execution, not just better inference.
  2 sacra 3 sacra 10 sacra

The likely endpoint is a smaller set of AI biotechs that look less like software vendors and more like industrialized drug companies. The winners will be the groups that can keep turning proprietary experiments into better models, then turn better models into real clinical assets faster than partners or rivals can.