Eli Lilly has switched on what it describes as the most powerful supercomputer ever owned and operated by a pharmaceutical company, and the implications for LillyPod drug discovery could reshape how medicines are found, tested, and delivered. Inaugurated in February 2026 at Lilly’s Indianapolis headquarters, LillyPod is built on NVIDIA’s DGX SuperPOD architecture and powered by 1,016 Blackwell Ultra GPUs. It delivers more than 9,000 petaflops of AI performance, the equivalent of nine quintillion mathematical operations every second. Assembled in just four months, the system represents a fundamental shift in how computational biology is being applied at industrial scale.
From Wet Lab to Dry Lab: Breaking the Physical Limit
For decades, the wet laboratory has been the bottleneck of pharmaceutical research. Scientists could test roughly 2,000 molecular ideas per drug target per year, a constraint imposed not by imagination, but by the physical limits of bench science. LillyPod changes that equation dramatically.
“Now the supercomputer centre essentially just breaks the physical limit of the wet lab,” said Yue Wang Webster, Vice President of Research and Development Informatics at Lilly. “Now in the dry lab, you can test billions of molecule ideas at your fingertips.”
The system creates what Lilly describes as a computational dry lab at massive scale, enabling scientists to simulate and evaluate billions of molecular hypotheses in parallel before committing to any physical experiments. Protein diffusion models, small-molecule graph neural network models, and genomics foundation models can all be trained and deployed within the same infrastructure.
Thomas Fuchs, senior vice president and chief AI officer at Lilly, framed the system not as IT infrastructure but as a scientific instrument.
“It’s like an enormous microscope for biologists,” he said. “Computation is at the heart of biology and it is at the heart of science. Being able to compute at scale is not something optional for a company like ours; it is absolutely necessary.”
What LillyPod Can Do
The scale of LillyPod’s capabilities is difficult to contextualise without a reference point. NVIDIA has noted that computational power once requiring seven million Cray supercomputers now fits inside a single Blackwell Ultra GPU and LillyPod contains more than 1,000 of them.
The system gives Lilly’s genomics teams access to 700 terabytes of data, supported by over 290 terabytes of high-bandwidth GPU memory. That infrastructure enables analysis of individual cell and molecule-level biology, exploration of vast chemical spaces, and the training of proprietary AI models on data generated from more than a billion dollars’ worth of internal research including, crucially, lessons learned from millions of molecules that failed.
Beyond molecule design, LillyPod supports medical imaging-based biomarker development, clinical trial design optimisation, and manufacturing efficiency improvements. Lilly’s production lines already use AI to photograph auto-injectors dozens of times in a fraction of a second, scanning for defects. The same computational layer now extends deeper into R&D.
Scientific AI agents, digital assistants capable of reasoning, planning experiments, and coordinating across physical and digital laboratories are being developed on the platform.
“AI agents can work 24/7 and explore ideas that humans might not have the time or capacity to experiment with,” said Diogo Rau, Executive Vice President and Chief Information and Digital Officer at Lilly.
TuneLab: Opening the Platform to the Broader Ecosystem
One of the more strategically significant aspects of the platform is what Lilly plans to do with the models it trains on it. A portion of those models will be made available through TuneLab, Lilly’s federated AI and machine learning drug discovery platform, which is positioned as a collaborative infrastructure for the wider biopharma community.
TuneLab operates on a federated learning framework built on NVIDIA FLARE, allowing partner companies to train models collaboratively while keeping their underlying proprietary data isolated. The platform is also set to incorporate NVIDIA BioNeMo open foundation models for healthcare and life sciences, making it one of the first drug discovery platforms to combine proprietary pharma models with open-source biological AI at scale.
Earlier in 2026, Lilly extended TuneLab access to external partners through collaborations with Benchling and Revvity. In 2025, selected biotech companies were offered access to TuneLab models at no cost in return for contributing data to further train Lilly’s AI systems, a strategic calculus that could allow Lilly to surface external compounds and targets it can later partner on or acquire.
The model is counterintuitive but deliberate. By seeding the ecosystem with better tools, Lilly benefits from a richer data network while expanding its influence across early-stage discovery. The platform represents a new kind of open innovation, where the infrastructure itself becomes a competitive moat.
What makes TuneLab technically distinctive is how the federated learning architecture actually works in practice. Rather than partners sending their data to Lilly, the process runs in reverse: the AI model travels to each partner’s own secure infrastructure, trains locally on their proprietary data, and returns only encrypted mathematical updates, not the underlying molecules, to a central server. Those updates are aggregated to improve the global model, which is then redistributed across all participants. No partner’s raw data ever leaves their own environment.
Lilly has deliberately focused TuneLab’s initial use cases on pre-competitive areas such as ADMET profiling (how a molecule is absorbed, distributed, metabolised, excreted, and whether it is toxic) and antibody developability, precisely because these are not areas where companies differentiate. The logic is straightforward: if everyone benefits from better shared tools in areas that don’t confer competitive advantage, more partners participate, more data flows into the system, and the models improve for all. According to Lilly’s head of TuneLab, the platform had onboarded over 70 biotech partners within its first few months, with a stated goal of reaching 150 by the end of 2026. In May 2026, Lilly extended TuneLab’s reach further still, announcing a partnership with Collaborative Drug Discovery to integrate TuneLab directly into CDD Vault, allowing biotech companies to run Lilly’s predictive models within their existing research workflows without any additional platform switching.
A $1 Billion Commitment and a Broader Investment Strategy
LillyPod is not a standalone initiative. It forms part of a broader five-year, $1 billion commitment by Lilly and NVIDIA to accelerate AI-driven drug discovery and integrate it across the company’s operations. A planned co-innovation hub in South San Francisco, designed to link wet-lab experimentation with large-scale computational modelling, is intended to generate high-quality data for training next-generation biology and chemistry foundation models.
That investment sits within an even larger strategic context. Lilly has committed $50 billion to expanding its US manufacturing and R&D footprint, which includes four new facilities and a proposed $4.5 billion laboratory in Indiana, the Lilly Medicine Foundry, focused on advanced manufacturing and drug development. The combined initiatives are expected to create approximately 13,000 high-wage manufacturing and construction jobs.
For a broader look at how the industry is betting on AI-powered platforms, see Agentic AI in Drug Discovery: Why Big Pharma is All in.
Managing Expectations: The Limits of Computational Biology
Despite the ambition, Lilly’s leadership has been careful to manage expectations around what AI can and cannot accelerate. Rau acknowledged the risk of overpromising.
“There’s a tendency to think that we’re now going to be able to discover new medicines in three months. That’s one that’s particularly damaging and destructive,” he said.
Fuchs has estimated it could be a decade before AI can generate, validate, and advance drug candidates end-to-end without significant human oversight. Biology, not compute, remains the bottleneck in clinical development. Cancer trials and neurological studies cannot be shortened by processing power alone; patient biology moves on its own timescale.
The more realistic near-term opportunity lies in the discovery phase, where LillyPod can dramatically widen the aperture of what is scientifically feasible to explore. Finding molecules that would never have been identified through conventional wet-lab approaches, and learning from the vast history of compounds that did not work, is where the system’s immediate value lies.
Sustainability and Responsible Deployment
Lilly has committed to running LillyPod on 100% renewable electricity by 2030, using efficient liquid cooling and minimising incremental energy impact. The system’s nearly 5,000 network connections are built with more than 450 kilograms of fibre cables. The infrastructure was designed with Lilly’s sustainability goal of carbon neutrality by 2030 in mind, using the company’s existing chilled-water systems for cooling.
The question of responsible AI deployment extends beyond energy. Lilly has described its approach as embedding AI as a core scientific capability rather than a standalone tool, with governance frameworks intended to ensure models are secure, ethically deployed, and compliant with healthcare regulations.
A New Era for Pharmaceutical Science
LillyPod represents a significant step in the industrialisation of drug discovery. Whether it delivers on its full potential will depend not just on compute, but on the quality of data fed into it, the rigour of the science surrounding it, and Lilly’s ability to translate computational hypotheses into clinical results over the years ahead. What is clear is that LillyPod drug discovery, at this scale and with this level of investment, sets a new benchmark for what a pharmaceutical company’s internal scientific infrastructure can look like.
“We are, right here, right now, at the right moment to advance biology in a way that has just never been done before,” said Rau at the system’s inauguration.
