LillyPod Goes Live: Pharma's Most Powerful AI Supercomputer

Summary: Eli Lilly brought LillyPod live at its Indianapolis campus on February 25, 2026, using 1,016 NVIDIA Blackwell Ultra GPUs in a DGX SuperPOD with AI performance above 9,000 petaflops. The system will support genomics, molecule design, imaging, and manufacturing workloads, while TuneLab uses federated learning over proprietary data. Lilly’s AI officer still calls biology, not compute, the bottleneck for clinical validation.

LillyPod is officially live: Eli Lilly cut the ribbon on the supercomputer at its Indianapolis campus on Wednesday, February 25, 2026. Inside sits a DGX SuperPOD configuration of 1,016 NVIDIA Blackwell Ultra GPUs; AI performance tops 9,000 petaflops. NVIDIA labels the installation the world’s largest and most powerful AI factory for drug discovery — and whichever yardstick you apply, nothing the pharmaceutical industry has fielded before comes close.

Seen as a timeline, the launch is a closing move rather than a one-off purchase. NVIDIA’s GTC Washington DC event hosted the first disclosure in October 2025; the timetable then called for a December buildout and a January go-live. January’s J.P. Morgan Healthcare Conference added the rest: a South San Francisco co-innovation lab run jointly with NVIDIA, plus a five-year partnership worth as much as $1 billion. Developers and product teams should watch the through-line here: a vertical-industry leader turning frontier compute into a platform it owns, end to end.

Inside the Box: 1,016 Blackwell Ultra GPUs

Begin with the raw numbers, which need no help. R&D World reports that no DGX SuperPOD of this configuration existed before; LillyPod is the first one assembled from DGX B300 systems. Lilly’s Chief AI Officer Thomas Fuchs supplied a comparison with more color: measured against the Cray systems Lilly was running in 1992, a single Blackwell Ultra GPU carries the compute of roughly 7 million of them — thirty-plus years of hardware evolution folded into one rack-scale unit.

Scale, though, is not the story; access is. Fuchs and Diogo Rau (Lilly’s chief information and digital officer) both stressed that Lilly’s scientists can now go straight to these GPUs — the resource is no longer rationed out by a small high-performance computing group but sits inside the daily workflow of bench researchers. What an AI factory ultimately proves depends on who uses it and how hard; a room full of idle GPUs demonstrates precisely nothing.

The Workload Lineup: Genomics Through Manufacturing

What Lilly intends to run on the machine is concrete: genomics; molecule and peptide design; single-cell biology; imaging; and even manufacturing operations. The economics beneath those workloads are blunt. A Lilly wet-lab team gets through roughly 2,000 hypotheses per target in a year; in silico, the figure becomes billions running at once. Compute changes the attempts-per-unit-time equation by orders of magnitude — the very reason drug discovery has spent the past several years adopting AI so aggressively.

TuneLab and Federated Learning: A Model Factory on Top of the Data

Hardware is half the story at most, and the software layer repays a closer look. TuneLab — Lilly’s AI/ML drug-discovery platform — builds its models on roughly $1 billion of the company’s proprietary data and runs federated learning over NVIDIA FLARE, so collaboration across institutions never demands pooling raw data. NVIDIA adds that TuneLab stands alone as the first drug-discovery platform carrying both Lilly’s own models and the Clara open foundation models from NVIDIA. Keeping proprietary models while staying plugged into the open ecosystem is the architectural decision most worth borrowing here.

Biology Is the Bottleneck: A Reality Check Measured in Decades

Step away from the ceremony and Fuchs was working to cool the excitement: “You cannot accelerate a cancer trial or a neuro trial — the bottleneck is biology.” His estimate for when AI can generate, validate, and advance drug candidates end to end: about a decade out. Anyone pricing “AI disrupts pharma” narratives gets a useful anchor here — compute scales exponentially, but clinical validation does not. An investor deck’s GPU count sits a long way from a molecule that has cleared the clinic, and the clock across that gap runs on biology.

Scientific AI Agents: What Comes Next

Lilly’s sights are now set on scientific AI agents — systems that can plan experiments, do reasoning, and line up work across computational and physical laboratories. In Fuchs’s framing, the job is to carry drug discovery out of its artisanal character and into an industrialized operation. CB Insights recently named Lilly the most AI-ready pharmaceutical company, and the physical assets keep attracting capital alongside the digital ones: a $50 billion commitment to U.S. manufacturing and R&D, the $4.5 billion Lilly Medicine Foundry, and job creation in the tens of thousands. Compute, data, and laboratories moving on three fronts at once — that is the complete reading of this news. What opened 2026 as an AI capex wave among cloud giants (our outlook from January) is now reaching the leaders of vertical industries.