Data Workflow and Drug Economics

The moat in AI oncology is being built at the point where data collection, doctor workflow, and drug ownership meet. A model alone can be copied or matched, but a company that sits inside the trial enrollment funnel, has longitudinal patient records tied to genomics, and also owns or co owns the drug program can decide which patients enter studies, learn faster from outcomes, and capture more of the economics if a drug works.

• Pathos is explicitly building around multimodal oncology data and predictive trial design, and its AstraZeneca and Tempus agreements were framed around creating a large oncology foundation model using Tempus data. That matters because Tempus already monetizes the data layer through licensing and model development fees, which shows where bargaining power sits in the stack.
  1 sacra 2 pathos 3 tempus
• Roche's Flatiron and Foundation Medicine show the incumbent version of this playbook. Flatiron contributes EHR derived real world data and trial operations software, Foundation contributes genomic profiling, and together they support evidence generation and trial design. In practice, this is less a software seat sale and more a pipeline for finding patients, defining biomarkers, and shaping studies.
  4 flatiron 5 foundationmedicine 6 flatiron 7 nature
• The real prize is downstream drug economics. A company that only sells AI tools gets paid fees, while a company that uses data and workflow control to advance its own assets can earn the far larger upside from drug approvals, licensing, and royalties. That is why Pathos is pairing platform partnerships with wholly owned development.
  1 sacra

This market is heading toward a small number of oncology stacks that combine patient data supply, embedded clinical workflow, and direct exposure to asset value. The winners will look less like model vendors and more like hybrid operators, part data network, part trial engine, and part biotech owner.