Nvidia, Eli Lilly commit $1B to AI drug discovery lab

Nvidia and pharmaceutical heavyweight Eli Lilly announced a joint investment of up to $1 billion over the next five years to establish a dedicated research laboratory focused on developing foundation models for AI-assisted drug discovery. The collaboration, revealed at the JPMorgan Healthcare conference, represents one of the largest single commitments to applying artificial intelligence to pharmaceutical research.

The co-innovation lab will be located in the San Francisco Bay Area and is scheduled to open later this year. It will physically bring together Eli Lilly's biologists and chemists with Nvidia's software engineers and model developers. This proximity aims to break down traditional silos between computational and experimental research teams.

Building a Continuous Discovery Pipeline

The lab's first major objective is to create what Nvidia describes as a "continuous learning system" that tightly connects Lilly's "agentic wet labs with computational dry labs." This system is designed to enable 24/7 experimentation, where computational resources continue working even when human researchers are away.

The concept mirrors continuous integration and continuous delivery (CI/CD) pipelines used in software development. When scientists leave for the evening, the compute infrastructure can automatically prepare datasets, run simulations, and optimize experimental parameters for the next day's physical lab work. This approach could significantly reduce the idle time between research phases.

Vera Rubin Platform Powers Foundation Models

The collaboration will leverage Nvidia's newly announced Vera Rubin compute platform, which the company unveiled at CES last week. Vera Rubin promises a fivefold performance increase over the current Blackwell GPU generation. These chips will provide the computational power necessary to train large-scale foundation models specifically tailored for biological and chemical research.

The lab is positioned to be among the first to access Vera Rubin hardware, though significant availability isn't expected until the second half of this year. In the interim, Eli Lilly already operates substantial compute infrastructure. The company deployed a Blackwell Ultra-based SuperPOD with 1,016 B300 GPUs at GTC DC last October, demonstrating their commitment to computational biology.

BioNeMo Framework Provides Software Foundation

The research will utilize Nvidia's BioNeMo platform, an open-source framework introduced in fall 2022 for building and training deep learning models specifically for drug discovery. BioNeMo provides pre-trained models for protein folding, molecular property prediction, and generative chemistry, which researchers can customize with proprietary data.

The platform supports various model architectures including transformers and diffusion models, which have shown promise in generating novel molecular structures with desired properties. By using BioNeMo, the lab can build upon existing foundations rather than starting from scratch.

Expanding Beyond Drug Discovery

While drug discovery remains the primary focus, the lab will also explore AI applications across the entire pharmaceutical value chain:

Clinical Development: AI models could help optimize trial design, identify suitable patient populations, and predict potential adverse effects before human trials begin.

Manufacturing Optimization: Eli Lilly is investigating Nvidia's Omniverse Robotics platform to optimize manufacturing plants and increase production capacity for high-demand drugs. This includes simulating production lines before physical implementation and using AI to identify bottlenecks.

Commercial Operations: The lab will explore how AI can improve supply chain management, demand forecasting, and regulatory submission processes.

Industry Context and Implications

This partnership reflects a broader trend of pharmaceutical companies moving beyond pilot projects to industrial-scale AI research. Traditional drug discovery takes 10-15 years and costs billions, with failure rates exceeding 90%. AI promises to accelerate molecule screening, predict toxicity earlier, and identify promising candidates that human researchers might miss.

However, challenges remain. Recent studies have highlighted limitations in AI's ability to perform complex biological reasoning, with some systems producing confident but incorrect predictions. The continuous feedback loop between computational and wet lab research aims to address these limitations by grounding AI models in empirical data.

The collaboration also demonstrates how hardware manufacturers like Nvidia are expanding beyond traditional computing markets into vertical industries. By embedding their technology directly into pharmaceutical workflows, they create ecosystems that lock in demand for their platforms.

Timeline and Next Steps

The lab will open later this year, with initial projects focusing on establishing the continuous learning infrastructure. Within the first 12-18 months, the teams expect to have trained initial foundation models on Lilly's proprietary chemical and biological datasets. By year three, the lab aims to have AI-generated drug candidates entering preclinical testing.

Success in this venture could establish a template for similar collaborations across the pharmaceutical industry, potentially reshaping how drugs are discovered and developed for decades to come.