Drug discovery has increasingly become a computing arms race, and Eli Lilly just staked a claim to the top of it. The pharmaceutical giant has brought online LillyPod, a supercomputing system built on NVIDIA’s DGX SuperPOD architecture that the company and NVIDIA describe as the largest AI system wholly owned and operated by a pharmaceutical company, according to reporting from HPCwire’s AIwire and RCR Wireless.
The hardware behind the claim
LillyPod is built around 1,016 NVIDIA Blackwell Ultra GPUs, delivering more than 9,000 petaflops of AI performance, according to coverage from AI Weekly and R&D World. The system was assembled in roughly four months after being first announced in October 2025 as part of Lilly’s broader “AI factory” initiative, and it was inaugurated in Indianapolis, where much of Lilly’s research and manufacturing operations are based.
What the system is actually for
Beyond raw processing power, LillyPod gives Lilly’s genomics and drug-discovery teams access to roughly 700 terabytes of data, backed by more than 290 terabytes of high-bandwidth GPU memory, according to the reporting. That scale of memory is intended to support training AI models at the resolution of individual cell and molecule biology, the kind of fine-grained modeling needed to predict how a candidate compound will behave in the human body long before it reaches a lab bench, let alone a clinical trial.
Part of a broader external-facing AI strategy
LillyPod is not purely an internal tool. Lilly has also built TuneLab, a platform that lets outside biotech partners tap into Lilly’s drug-discovery models using their own proprietary data, without that data ever leaving the partner’s control or being shared back to Lilly. TuneLab draws on more than $1 billion in Lilly’s own proprietary datasets, and the company says it aims to have 150 biotech partners enrolled by the end of 2026, with more than 70 already on board as of the latest reporting.
Why the industry is watching closely
Pharmaceutical companies have historically relied on cloud computing partnerships or academic supercomputing centers rather than owning AI infrastructure of this scale outright. Lilly’s decision to build and operate LillyPod itself, rather than lease equivalent capacity from a hyperscaler, signals a bet that owning the compute layer is now a strategic necessity for staying competitive in AI-driven drug discovery, not just a cost center to be outsourced. Skeptics note that raw compute capacity does not by itself guarantee better drugs; the harder, unresolved problem in AI drug discovery remains translating in silico predictions into compounds that succeed in expensive, failure-prone human clinical trials, where most candidates still fail regardless of how they were designed.
Part of an industry-wide computing buildout
Lilly’s move is one of the more visible examples of a broader 2026 trend of pharmaceutical companies investing directly in dedicated AI infrastructure rather than relying solely on cloud-provider partnerships. Industry surveys have found roughly 80% of biotech and pharma organizations plan to increase their AI budgets over the next 12 months, with about 23% expecting to double spending or more, a shift analysts have described as the industry moving from isolated pilot tools toward fully integrated, AI-native discovery systems. Against that backdrop, LillyPod functions as both a research tool and a public statement of intent, signaling to investors, partners and rivals alike that Lilly intends to compete on the scale of its own AI infrastructure rather than simply licensing capacity from outside vendors.
What’s next
The real test for LillyPod will not be its petaflop rating but whether Lilly’s pipeline starts moving faster through discovery and into the clinic as a result, and whether TuneLab succeeds in pulling in the 150 external biotech partners the company is targeting by year end. If the platform delivers, expect rival pharmaceutical giants to follow with their own dedicated AI supercomputing buildouts rather than continuing to rely solely on public cloud infrastructure for large-scale drug discovery modeling.
The power and cost side of the equation
Systems built around more than a thousand of NVIDIA’s Blackwell Ultra GPUs, the generation of chips NVIDIA began shipping in 2025, draw substantial electricity and require dedicated cooling infrastructure, costs that have become a growing point of scrutiny across the AI industry as data-center power demand strains local grids in several U.S. regions. Lilly has not disclosed LillyPod’s specific power consumption or the total capital cost of the buildout, but industry estimates for comparable DGX SuperPOD-class installations of this scale typically run into the hundreds of millions of dollars once facility, networking, and cooling costs are included alongside the GPUs themselves, underscoring that LillyPod represents a multi-year capital commitment rather than a one-time hardware purchase.
