Eli Lilly's AI Factory Goes Live With 1,016 Blackwell GPUs

Drug companies have spent decades accumulating proprietary experimental data: hundreds of thousands of molecules tested, safety profiles logged, preclinical results catalogued. Eli Lilly is betting that data can now be the engine of a genuine AI advantage, one built on owned hardware rather than rented cloud.

In February 2026, Lilly inaugurated LillyPod at its Indianapolis headquarters, an on-premises AI factory powered by 1,016 Nvidia Blackwell Ultra GPUs that delivers more than 9,000 petaflops of AI performance, according to Nvidia. The system was assembled in four months and gives Lilly's genomics teams access to 700 terabytes of data backed by over 290 terabytes of high-bandwidth GPU memory, enough headroom to train models at the level of individual cell and molecule biology.

The external layer of the strategy is TuneLab, covered by the Financial Times: a federated AI and machine learning platform that lets biotech companies run Lilly's drug discovery models against their own data without sharing it. The first release includes 18 models, including ones that predict small molecule properties and assess antibodies, trained on proprietary datasets Lilly values at over $1 billion. More than 70 biotech partners have already signed on, with Lilly targeting 150 by end of 2026. In return for access, selected biotech partners contribute training data, which feeds back into continuous improvement of the shared models.

The federated design is the structurally interesting part. It lets Lilly build a network that accumulates value from partner data contributions without requiring those partners to hand over their IP. Whether that distinction holds as the platform scales, and whether regulators or partners start asking harder questions about the training feedback loop, is worth watching.

What the reporting does not give you is clinical evidence: there is no public account of how much faster any specific drug candidate has moved through Lilly's pipeline since LillyPod came online. Drug development timelines run years, so definitive ROI from this infrastructure is probably some distance away. The clearer near-term beneficiaries are the biotech startups gaining access to AI models they could not build independently, trained on data priced at a scale that most smaller companies will never accumulate.