1. What happened
Recursion receives limited licensed access to Tempus's proprietary de-identified clinical and molecular data for therapeutic product development, including model training, improvement, modification and derivative works. Downloaded records are subject to per-time and aggregate unique-record caps and historically a 180-day retention limit. The September 2026 amendment extends the term to six years, removes convenience termination and lowers the remaining unique-record cap.
2. Buyer / seller
Recursion Pharmaceuticals is the AI buyer or model-development side. Tempus AI owns or contributes the proprietary corpus.
3. Economics
The 2026 amendment sets $14 million annual license fees on each of the third, fourth and fifth anniversaries, at least $4 million of each payable in cash and the remainder in cash and/or Recursion shares. The original 2023 deal disclosed up to $160 million over five years. Economics scope: $42 million of amended third-through-fifth anniversary license fees is explicitly disclosed; amounts already paid under the original schedule and the economic effect of the reduced unique-record cap are not restated here.. Strategic Signal has not estimated undisclosed consideration.
4. Proprietary data
Tempus's proprietary de-identified patient-centric multimodal oncology database of clinical and molecular records, originally described by Recursion as more than 20 petabytes, licensed for therapeutic-development AI/ML use subject to unique-record caps and retention limits. Historical depth: Tempus's oncology database aggregates clinical and molecular patient records accumulated over time; the exact date range and longitudinal completeness are not disclosed in the amendment.. Refreshability: The arrangement spans six years and grants continuing licensed access subject to aggregate unique-record limits; the exact refresh cadence and amended unique-record count are not disclosed..
5. Rights and restrictions
Use is restricted to therapeutic product development within a secure Recursion environment and subject to download, aggregate unique-record and retention controls. Ownership: No ownership transfer. Tempus retains ownership of Licensed Data and Tempus Materials.. Training: Permitted for Recursion and affiliates to develop and train AI/ML models solely for therapeutic product development, subject to the agreement's limits.. Post-training: Permitted to improve and modify Recursion AI/ML models and develop embeddings, training and test data, subject to contractual restrictions..
6. Why the data matters for AI
Training, improving, modifying and creating derivative works of Recursion causal AI/ML models for therapeutic product development, plus biomarker and patient-stratification work. The corpus supports therapeutic hypotheses, biomarker strategies, patient stratification and model learning from linked real-world clinical and molecular observations.
7. Asset Score — 94/100
Strategic Signal analysis. Tempus's large proprietary multimodal oncology corpus combines real-world clinical and molecular observations with explicit AI-training utility and strong therapeutic decision relevance. Score coverage: 100%.
| Factor | Weight | Rating / 5 | Points |
|---|---|---|---|
| Uniqueness / scarcity | 12 | 4.5 | 10.8 |
| Historical depth | 8 | 4.5 | 7.2 |
| Decision → outcome richness | 13 | 4.5 | 11.7 |
| Decision Value Density | 13 | 5 | 13.0 |
| Real-world grounding | 10 | 5 | 10.0 |
| Domain economic value | 9 | 5 | 9.0 |
| Refreshability | 8 | 4.5 | 7.2 |
| Proprietary advantage | 9 | 4.5 | 8.1 |
| Model-learning usefulness | 10 | 5 | 10.0 |
| Non-replicability | 4 | 4 | 3.2 |
| Rights usability | 4 | 4.5 | 3.6 |
8. Transaction Signal Score — 96/100
Strategic Signal analysis. The agreement has unusually explicit model-use rights and separately disclosed multi-year license economics, strengthened by a filed 2026 amendment; no competitive process or independent uplift study is disclosed. Score coverage: 85%.
| Factor | Weight | Rating / 5 | Points |
|---|---|---|---|
| Disclosed economics | 16 | 5 | 16.0 |
| Clean price discovery | 16 | 4 | 12.8 |
| Separability of data consideration | 15 | 5 | 15.0 |
| Explicit training / model-improvement use | 14 | 5 | 14.0 |
| Clarity of rights purchased | 12 | 5 | 12.0 |
| Competing bids | 10 | Unknown | — |
| Strategic buyer quality | 6 | 4.5 | 5.4 |
| Repeatability / precedent value | 6 | 5 | 6.0 |
| Independent model-uplift evidence | 5 | Unknown | — |
9. M&A / valuation implications
Demonstrates that a proprietary healthcare-data asset can command large, separately stated multi-year licensing economics when explicit AI-training and derivative-model rights are granted without selling the underlying corpus.
10. Evidence boundary
Primary SEC filings establish the proprietary corpus, explicit AI/ML training and derivative-work rights, original and amended economics, six-year term, removal of convenience termination, lower aggregate unique-record cap, ownership boundaries and de-identification. The amended exact record cap, exclusivity, current refresh cadence, geographic restrictions and independent model-uplift evidence are not disclosed.