Big Pharma’s Billion-Cell Gamble on AI

Illumina’s Billion Cell Atlas teams with Lilly, AstraZeneca, and Merck to reshape how AI learns human biology

13 Jan 2026

The pharmaceutical industry’s next advance may owe less to chemistry than to cartography. Instead of hunting for clever new molecules, drugmakers are trying to map biology itself, cell by cell, at industrial scale, and let machines learn from it.

That is the bet behind Illumina’s Billion Cell Atlas, a collaboration with Eli Lilly, AstraZeneca and Merck. The project aims to record how vast numbers of human cells behave when particular genes are switched on or off. The resulting dataset would be among the largest ever assembled in biomedicine. Its purpose is simple: to give artificial intelligence something worth learning from.

For years drug firms have touted AI as a way to speed up discovery and cut costs. The results have disappointed. Algorithms trained on small, messy or inconsistent datasets have often produced ideas that looked clever but failed in the lab or clinic. The problem, many now concede, was not the models. It was the data.

As one researcher close to the project puts it, the goal is to feed AI biology it can actually understand. Capturing how cells respond to genetic change across huge populations should reveal patterns that are invisible in smaller experiments. Those patterns matter most early in research, when targets are chosen and mistakes are cheap.

Illumina supplies the sequencing machinery that makes such scale feasible. The firm argues that computing power is no longer the main constraint on AI in life sciences. What holds progress back is the lack of large, standardised datasets that reflect how cells really behave. By focusing on consistency as well as volume, the atlas seeks to fix that.

For Lilly, AstraZeneca and Merck, the appeal is practical. Better insight into disease mechanisms before clinical trials begin could help them back stronger targets and avoid expensive failures later on. In an industry where late-stage disappointments are common and costly, even modest improvements upstream can pay handsomely.

The project also points to a wider shift. Drug discovery has long been secretive, with data hoarded inside firms. Building shared infrastructure suggests a grudging recognition that some problems are too big to solve alone. Questions remain about who gets access and whether smaller biotech firms will benefit. But the direction is clear.

If biology can be treated as a scalable data problem, AI may yet earn its keep. Alliances like this one will help decide which firms are best placed to do so.