Pharmaceutical development has long been one of the toughest challenges for AI — complex data, strict regulations, high failure rates, and extremely long cycles.
But in February, Eli Lilly unveiled a piece of equipment that made the industry take notice: LillyPod.
What is LillyPod
LillyPod is an AI supercomputing cluster built, owned, and operated entirely by a single pharmaceutical company — a first for the industry.
Hardware configuration: 1,016 NVIDIA Blackwell Ultra (DGX B300) GPUs, delivering a total computing power of over 9,000 petaflops (9 quintillion floating-point operations per second).
From decision to deployment, it took just four months.
Lilly positions it as an AI factory — not for a single task, but to run the entire drug development pipeline: from protein structure prediction and genomic analysis to compound screening, clinical trial design, and manufacturing optimization.
Why Lilly Built It In-House
Traditional pharmaceutical companies typically outsource AI to cloud providers or AI service vendors. Lilly chose to keep computing power under its own control.
Several reasons:
- Drug development data is extremely sensitive; external clouds pose data sovereignty and compliance risks
- In-house infrastructure allows specialized optimization for pharmaceutical workflows without competing for resources with other customers
- 700 TB of genomic data and 290 TB of GPU high-bandwidth memory — running at this scale externally would be prohibitively expensive
The $1 Billion Partnership with NVIDIA
LillyPod is just one part of the Lilly-NVIDIA collaboration.
The two companies also announced they will jointly establish an AI Co-Innovation Lab in San Francisco, with a combined investment of up to $1 billion over five years, covering talent, infrastructure, and computing power.
Key focus areas: protein diffusion model training, small-molecule graph neural networks, and genomic foundation models.
This effectively brings AI into the earliest and most expensive stages of new drug development — also the stages with the highest failure rates.
Industry Context
In 2026, AI-driven drug discovery has moved from promising to delivering results.
Several concurrent developments:
- Eli Lilly acquired several preclinical candidate compounds from Insilico Medicine (an AI drug discovery company) for up to $2.75 billion
- A Chinese research team introduced the DrugCLIP framework, which can scan 10 million compound-target pairs in hours — 10 million times faster than traditional virtual screening
- Multiple AI-designed drug candidates are advancing through clinical trials
Pharmaceutical companies are finding that the problem is no longer finding candidate molecules, but quickly selecting the most promising ones for clinical trials from millions of candidates.
What LillyPod aims to do is remove computing power as a constraint on that selection process.
Sources: CocoLoop, Lilly Launches LillyPod NVIDIA DGX SuperPOD for Genomics and Drug Discovery AI (HPCwire/AIwire); NVIDIA and Lilly Announce Co-Innovation AI Lab to Reinvent Drug Discovery (NVIDIA Newsroom); AI in Drug Discovery Hits New Milestone (AI News International)