GSK Paid $110 Million for a Dataset. Most Companies Don't Know Theirs Is Worth Anything at All.

On 30 July 2026, GSK announced a deal with a British AI biotech called Relation Therapeutics worth up to $110 million. However, there was no drug, no clinical candidate and no molecule at any stage of development in the deal. What GSK agreed to pay for was the data, which was a proprietary biological dataset that Relation had been building for years and the capability to generate significantly more of it.

Relation Therapeutics is a London-based AI biotech founded in 2020. Its core focus has been building computational models of how human cells respond to genetic and pharmacological interventions, essentially, teaching AI to predict the behaviour of living biology. To do that, it has spent years generating large-scale human cellular perturbation data in automated laboratories: experiments that probe how cells across different disease contexts respond when specific genes are switched on or off, or when drug-like compounds are introduced. That data, accumulated methodically over several years of research, became the foundation for MORGAN (Multi-Omic Regulatory Genomics using Artificial Neural Networks), Relation's flagship AI foundation model of cellular response.

This is a real, proprietary biological dataset that Relation had built as a byproduct of doing its science and that GSK recognised as one of the most valuable inputs it could acquire for its own AI drug discovery programme.

To understand why GSK valued it so highly, it helps to understand what is actually scarce in AI-driven drug discovery right now. Foundation models for biological prediction have proliferated rapidly and the computational architectures underlying them are increasingly well understood and widely available. What remains scarce (and what is becoming more valuable as a result) is the high-quality, proprietary, human-derived biological data needed to train those models to a standard that produces reliable predictions. Public datasets can be too small, too narrow or too inconsistent. The organisations that have accumulated large, well-structured, disease-relevant biological datasets through years of careful research have something that cannot be quickly replicated regardless of budget.

In the same period, GSK signed a multi-year partnership with Helix for large-scale genomic cohort data, paid $50 million to Noetik for spatial oncology virtual cell models built on proprietary training datasets and integrated Microsoft Discovery into its AI workflows. The strategy is coherent: GSK has identified proprietary biological data as the primary competitive input it needs to acquire, and it is building a portfolio of it systematically. The Relation deal is the latest and largest expression of that strategy.

The broader implication of this deal extends well beyond pharmaceutical research. Most businesses that hold proprietary datasets accumulated over years of operation have never had those assets independently assessed or valued. They exist as operational byproducts, the output of running the business, rather than as recognised financial assets. But the logic that made Relation's dataset worth $110 million to GSK applies across sectors. A dataset that is specific to a particular domain, accumulated over many years, built on proprietary methodology and relevant to a buyer who cannot easily replicate it elsewhere has commercial value that is almost never reflected anywhere in the owning company's financial statements.

The gap between what Relation knew it had built and what GSK was willing to pay for it is the gap that most data-rich businesses are sitting in right now. The companies that understand this earliest are the ones that will capture that value rather than leaving it on the table.

The question for every business sitting on years of proprietary, accumulated, sector-specific data is whether they understand what theirs is worth before someone else works it out for them.

Data Valuation Partners helps businesses understand what their data assets are worth. If you would like to explore what yours might be, get in touch.