Discovery Loop with CROs and Biofoundries
Discovery Loop
The key strategic move is to use outside lab capacity as the bridge from software to real world discovery. In biology and chemistry, the hard part is not generating candidate molecules on a computer, it is testing thousands of them in a repeatable way. CROs and biofoundries already run standardized assays, liquid handling, and high throughput screens, so a platform like Discovery Loop can plug its design engine into existing wet lab capacity instead of building its own labs first.
-
Tahoe shows what this can look like in practice. Its Mosaic platform pooled many disease models, ran large scale drug perturbation experiments, and produced a 100M cell atlas for training virtual cell models. The lesson is that better closed loop biology starts with owning the experiment design and data layer, even if parts of execution are heavily automated and standardized.
-
CROs matter because they already sell assay development, screening, and validation as a service. That means Discovery Loop could send a ranked batch of protein or molecule designs to an external lab, get structured assay results back, and feed those results into the next model iteration. The customer buys faster cycles, not just model output.
-
This pattern also fits materials and energy. The DOE Genesis Mission is explicitly pushing AI driven autonomous laboratories and partnerships that connect models, experimental facilities, and data. That creates a path for software companies to win around orchestration, data feedback, and facility access before they own any physical infrastructure themselves.
Over time, the winners in scientific AI will look less like pure model vendors and more like operating systems for experimentation. The company that can route designs into labs, capture clean results, and improve every cycle will build the most valuable asset, which is proprietary outcome data that makes each next discovery loop faster and more accurate.