On Wednesday, February 25, 2026, Eli Lilly held a ribbon-cutting at its Indianapolis campus to mark LillyPod going live. The supercomputer packs 1,016 NVIDIA Blackwell Ultra GPUs into a DGX SuperPOD configuration delivering more than 9,000 petaflops of AI performance. NVIDIA calls it the world’s largest and most powerful AI factory for drug discovery; by any measure, it is the most powerful supercomputer the pharmaceutical industry has ever built.
The launch is less a one-off purchase than the closing move of a longer play. Lilly first disclosed the system in October 2025 at NVIDIA’s GTC Washington DC event, with the buildout slated for December and systems online in January. At the J.P. Morgan Healthcare Conference in January, the company added a co-innovation lab with NVIDIA in South San Francisco and a partnership worth up to $1 billion over five years. For developers and product teams, LillyPod is a full-length case study in how a vertical-industry leader turns frontier compute into an internal platform.
The Spec Sheet: 1,016 Blackwell Ultra GPUs
The numbers speak for themselves. R&D World reports this is the first DGX SuperPOD of its configuration, built from DGX B300 systems. Lilly’s Chief AI Officer Thomas Fuchs offered a more vivid comparison: a single Blackwell Ultra GPU delivers roughly the compute of 7 million of the Cray systems Lilly was using in 1992. Three decades of hardware evolution, compressed into one rack-scale unit.
But scale is not the point — access is. Fuchs and Diogo Rau, Lilly’s chief information and digital officer, both emphasized that Lilly scientists can now reach these GPUs directly, moving compute out of the hands of a small high-performance computing group and into the daily workflow of bench researchers. An AI factory is only worth what its utilization and its user base say it is; a big room of idle GPUs proves nothing.
From Genomics to Manufacturing: What Runs on It
The target workloads are concrete: genomics, molecule and peptide design, single-cell biology, imaging, and even manufacturing operations. The economics behind them are blunt. Lilly’s wet-lab teams can test on the order of 2,000 hypotheses per target per year. In silico, billions of hypotheses run in parallel. Compute rewrites the number of attempts per unit time by orders of magnitude — which is exactly why drug discovery has been so aggressive in adopting AI over the past several years.
TuneLab and Federated Learning: Turning Data into a Model Factory
The hardware is only half the story; the software layer deserves closer attention. TuneLab, Lilly’s AI/ML drug-discovery platform, trains models on roughly $1 billion worth of Lilly’s proprietary data and uses federated learning on NVIDIA FLARE, so cross-institution collaboration never requires pooling raw data. NVIDIA also notes TuneLab is the first drug-discovery platform to offer both Lilly’s in-house models and NVIDIA’s Clara open foundation models. Building your own models while staying plugged into the open ecosystem is the architectural move most worth copying here.
Compute Isn’t the Bottleneck: A Decade-Sized Reality Check
Away from the ceremony, Fuchs deliberately poured cold water on the excitement: “You cannot accelerate a cancer trial or a neuro trial — the bottleneck is biology.” He estimates it will take about a decade before AI can generate, validate, and advance drug candidates end to end. For anyone pricing “AI disrupts pharma” narratives, that is a useful anchor: compute can grow exponentially; clinical validation cannot. There is a long distance between a GPU count in an investor deck and a molecule that has actually cleared the clinic, and the clock in between runs on biology.
Scientific AI Agents: The Next Milestone
Lilly’s next target is scientific AI agents — systems that plan experiments, reason, and coordinate across computational and physical laboratories. In Fuchs’s framing, the goal is to move drug discovery from an artisanal process to an industrialized one. CB Insights recently ranked Lilly the most AI-ready pharmaceutical company, and the firm keeps pouring capital into physical assets alongside the digital ones: a $50 billion U.S. manufacturing and R&D commitment, the $4.5 billion Lilly Medicine Foundry, and tens of thousands of jobs. Compute, data, and laboratories advancing on three fronts is the complete reading of this news. The AI capex wave that opened 2026 (our outlook from January) is spreading from cloud giants to the leaders of vertical industries.
Sources
- Eli Lilly’s LillyPod supercomputer goes live with 1,016 NVIDIA Blackwell GPUs — R&D World
- Lilly Deploys World’s Largest, Most Powerful AI Factory for Drug Discovery Using NVIDIA Blackwell-Based DGX SuperPOD — NVIDIA Blog
- Lilly Launches LillyPod NVIDIA DGX SuperPOD for Genomics and Drug Discovery AI — HPCwire
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.