Data moats are real but type-dependent — phenomics and clinical data are defensible; computational chemistry moats are eroding

Confidence: 7/10 Supporting evidence:

• Tempus AI 126% NRR (2025) — clinical data monetization is working. Evidence: strong
• Recursion's 2.5T phenomics images are 10-100x any academic equivalent and took 10+ years to build. Evidence: strong
• ESM3 open-weight release commoditizes the protein foundation model layer. Evidence: strong

Challenging evidence:

• Tempus' NRR could reflect novelty of AI tools rather than data lock-in specifically. Evidence: weak
• Public databases (PDB, ChEMBL, UniProt) are growing faster than ever; long-run data advantage may erode. Evidence: moderate

Evolution:

• Jun 2026 — Initial thesis at 7/10; high conviction on type-dependence

Depends on: data-moat Would change if: A major AI drug company achieves Phase 2 success using entirely public data; or Tempus NRR drops below 110%

---