Lilly & NVIDIA Launch AI Factory with Blackwell for Drug Discovery | NVIDIA Blog

Eli Lilly and Company this week launched LillyPod, what the company calls the world’s most powerful AI factory wholly owned and operated by a pharmaceutical company. The latest infrastructure, built in just four months, aims to accelerate medical advancements through large-scale data analysis and AI model training, representing a significant investment in computational power for drug discovery, and development.

A Leap in Computational Scale

LillyPod is centered around an NVIDIA DGX SuperPOD, equipped with 1,016 NVIDIA Blackwell Ultra GPUs, delivering over 9,000 petaflops of AI performance. To put that into perspective, NVIDIA notes that the computational power once requiring 7 million Cray supercomputers now resides within a single NVIDIA GPU. This dramatic increase in processing capability allows Lilly’s genomics team to work with 700 terabytes of data, utilizing over 290 terabytes of high-bandwidth GPU memory. The scale of the infrastructure is physically impressive as well, with nearly 5,000 connections built using more than 1,000 pounds of fiber cables.

“It’s a big day for us with the supercomputer coming on board, but it’s a day 150 years in the making,” said Diogo Rau, executive vice president and chief information and digital officer at Lilly. “LillyPod is a powerful symbol of who we are and why we do this work: to make life better for people around the world. We are, right here, right now, at the right moment to advance biology in a way that has just never been done before.”

How LillyPod Will Advance Drug Discovery

The core aim of LillyPod is to overcome the traditional constraints of pharmaceutical research, particularly the limitations imposed by “wet lab” experiments. Historically, drug discovery has been a slow process, with teams typically analyzing around 2,000 molecular ideas per target annually due to the time and resources required for physical synthesis and testing. LillyPod aims to dramatically increase this throughput by creating a “computational dry lab” where billions of molecular hypotheses can be simulated and evaluated in parallel. As Yue Wang Webster, vice president of research and development informatics at Lilly, explained, the supercomputer essentially “breaks the physical limit” of traditional research.

Specifically, LillyPod will support the training of several types of advanced AI models, including protein diffusion models, small-molecule graph neural networks, and genomics foundation models. These models will be applied across the entire pharmaceutical value chain, from initial target identification to clinical trial design and manufacturing optimization. The company intends to leverage these capabilities to design better trials, optimize production processes, and accelerate decision-making.

A Secure and Scalable Platform

NVIDIA’s full-stack AI factory architecture, including accelerated computing, NVIDIA Spectrum-X Ethernet networking, and optimized AI software, provides a secure and scalable platform for the highly regulated healthcare and life sciences industries. NVIDIA Mission Control software will be used to manage the DGX SuperPOD, orchestrate workloads, monitor performance, and automate AI operations. Here’s crucial for maintaining data integrity and compliance with stringent regulatory requirements.

Expanding Access Through Lilly TuneLab

Lilly isn’t keeping the benefits of LillyPod entirely to itself. Select models developed using the supercomputer will be made available through Lilly TuneLab, an AI and machine learning platform. TuneLab provides biotech companies with access to drug discovery models built on proprietary Lilly data, which represents an investment of over $1 billion. This platform utilizes a federated learning infrastructure built on NVIDIA FLARE, allowing companies to tap into powerful AI models while maintaining the privacy and security of their own data. As more companies participate, the models are expected to improve, creating a collaborative ecosystem for AI-driven drug discovery.

Beyond Drug Discovery: Agentic AI and Internal Innovation

The capabilities of LillyPod extend beyond traditional model training. Lilly employees can also use the infrastructure to build chatbots, agentic workflows, and research lab agents, reducing the necessitate to develop these tools from scratch. This internal innovation potential is a key benefit of having such a powerful computational resource in-house. Thomas Fuchs, senior vice president and chief AI officer at Lilly, emphasized the importance of computation in modern biology, stating that it’s “absolutely necessary” for a company like Lilly to build the computational future of medicine.

Sustainability Considerations

Lilly has committed to running LillyPod on 100% renewable electricity by 2030. The supercomputer utilizes efficient liquid cooling to minimize energy consumption and reduce its environmental impact. This commitment to sustainability reflects a growing awareness of the energy demands of large-scale AI infrastructure.

Looking Ahead: Collaboration and the Future of AI in Pharma

Lilly and NVIDIA are collaborating closely on this initiative, and will be presenting further details at NVIDIA GTC. The launch of LillyPod signals a broader trend of pharmaceutical companies investing heavily in AI and machine learning to accelerate drug discovery and development. The combination of scientific expertise, vast datasets, and powerful computational resources promises to unlock new possibilities in the fight against disease. As Fuchs succinctly put it, “This machine is exactly how AI should be used – it should be used for science. It should be used to lessen suffering and improve the human condition.”