Autonomous Labs Create Data Moats
Discovery Loop
This is a moat built from owning the experiment, not just the model. Lila and Periodic do not just read papers and simulate outcomes, they run instruments, collect samples, measure results, and feed those fresh observations back into training. That matters because published science is thin, delayed, and biased toward positive results, while real lab output includes failed runs, edge cases, and instrument level signals that improve automation in ways a software only rival cannot cheaply copy.
-
Lila pairs a scientific reasoning model with AI Science Factories and autonomous RNA workflows. Each cycle creates new wet lab data from its own robots and assays, so its model learns from proprietary experimental traces rather than only public biology datasets.
-
Periodic is applying the same pattern in materials science. Its pitch is not just a large model, but a model deployed inside physical labs for XRD and related analysis, where lab generated measurements become training data tied to real instruments and workflows.
-
A software only platform like Discovery Loop can move fast in simulation and code, but physical domains are messier. Lab hardware, sample prep, assay settings, and measurement noise vary by workflow, so the company that owns execution captures the best feedback loop and the hardest data to reproduce.
The next step is a split market. Horizontal software will remain strongest where discovery can stay inside compute, while the highest value biology and materials programs will reward companies that combine models with instrument control and proprietary data generation. Over time, the best science models are likely to look less like chatbots and more like operating systems for autonomous labs.