Kurz gesagt: In 2025 the largest pharmaceutical companies moved from buying artificial intelligence (AI) to building it on someone else’s data. Tempus signed agreements worth $200 million to license its oncology data and co-develop a foundation model with AstraZeneca and Pathos, and Bayer tapped ConcertAI’s database of millions of cancer records. The durable asset in oncology AI is the curated, de-identified patient data, and whoever owns it sets the terms.
The deals that defined the shift
Two agreements made the pattern clear. Tempus signed expanded agreements with AstraZeneca and Pathos to build what the partners call the largest multimodal foundation model in oncology, with Tempus receiving roughly $200 million in data-licensing and model-development fees. Tempus brings the raw material: more than 7.3 million de-identified patient records spanning clinical notes, genomics, imaging, and transcriptomics.
Days apart, ConcertAI announced a multi-year agreement with Bayer to put its Translational360 platform to work in precision oncology. That platform draws on a research-grade clinical and molecular database tied to a network of de-identified records from more than 9 million cancer patients across the United States.
The model commoditizes, the data compounds
These deals signal a shift from adopting AI models to building them. As model architectures spread and open weights proliferate, the algorithm becomes the commodity. The scarce input is regulatory-grade, consented, multimodal patient data at scale, cleaned and linked across genomics, imaging, and outcomes.
That asset takes years and deep clinical relationships to assemble, which makes it defensible. A pharma major can license a model in an afternoon, yet replicating a curated repository of millions of linked oncology records remains hard. The company that holds the data holds the leverage, and the licensing fees follow.
Why the structure favours data owners
The economics reward the data holder twice. The first payment arrives as a licensing or development fee, as the Tempus agreements show. The second accrues over time, because the same dataset can power model after model across many partners, each paying for access. A great dataset behaves like an annuity, paying out year after year.
The arrangement also keeps the data owner close to the science. By co-developing the model, the data partner gains insight into drug targets and trial design, which opens further commercial paths. Ownership of the input turns into a seat at the table for the output.
What it means for investors and founders
The read-through for backers is direct. In oncology AI, the company worth funding owns proprietary, consented, multimodal clinical data and the rights to license it. The defensibility lives in that data, since competitors can readily match a model. Diligence should focus on data provenance, consent scope, breadth across modalities, and the contracts that govern reuse.
That lens sits at the core of how Allegory Capital evaluates health AI. The most investable companies treat their data as the product, build the rights and infrastructure to license it cleanly, and become the partner that every drugmaker training a model will need. The 2025 deals are early proof that the gold in oncology AI sits in the data, and that the market will pay for it.
Referenzen
- Tempus in line for $200M from AstraZeneca, Pathos deal to develop cancer model: https://www.fiercebiotech.com/biotech/tempus-ai-line-200m-astrazeneca-pathos-deal-develop-cancer-model
- Tempus signs expanded strategic agreements with AstraZeneca and Pathos for the largest multimodal foundation model in oncology: https://www.tempus.com/news/tempus-signs-expanded-strategic-agreements-with-astrazeneca-and-pathos-to-develop-the-largest-multimodal-foundation-model-in-oncology/
- ConcertAI announces strategic agreement with Bayer to accelerate clinical development in precision oncology: https://www.businesswire.com/news/home/20250422551624/en/ConcertAI-Announces-Strategic-Agreement-with-Bayer-to-Accelerate-Clinical-Development-in-Precision-Oncology
