Lilly Launches Blackwell-Powered AI Factory With 1,016 GPUs for Drug Discovery

From pilot AI projects to dedicated pharmaceutical compute

According to NVIDIA’s February 26, 2026 report, Lilly has launched LillyPod, described as the most powerful AI factory wholly owned and operated by a pharmaceutical company. NVIDIA says the system is the first DGX SuperPOD with DGX B300 systems and includes 1,016 Blackwell Ultra GPUs, delivering more than 9,000 petaflops of AI performance. The build was assembled in four months and inaugurated in Indianapolis.

The strategic significance is less about headline FLOPS and more about control. Instead of treating advanced AI as an external cloud service layer, Lilly is positioning compute as an internal scientific instrument that can be tuned directly to regulated drug-development workflows.

What the system is designed to do

NVIDIA says LillyPod will support large-scale training of protein diffusion models, small-molecule graph neural network models, and genomics foundation models. The post also states that Lilly’s genomics teams can work with roughly 700 terabytes of data and over 290 terabytes of high-bandwidth GPU memory. Nearly 5,000 system connections are linked with more than 1,000 pounds of fiber cabling.

The practical goal is to expand pre-lab exploration. Lilly executives cited a historical constraint where wet-lab teams can typically test around 2,000 molecular ideas per target per year. With a large computational dry lab, teams can evaluate billions of molecular hypotheses in parallel before selecting physical experiments.

TuneLab, federated learning, and ecosystem effects

The post says selected models will be exposed through Lilly TuneLab, where biotech partners can access drug-discovery models built on proprietary Lilly datasets developed at more than $1 billion in cost. Lilly also plans to combine proprietary models with NVIDIA BioNeMo open foundation models, using a federated setup based on NVIDIA FLARE so participant data remains isolated.

Lilly additionally targets 100% renewable electricity for the infrastructure by 2030, alongside liquid cooling. If the platform performs as intended, the downstream impact could extend beyond discovery speed into trial design quality, manufacturing optimization, and the economics of translating model outputs into real medicines.