Lilly’s TuneLab platform and new AI supercomputing infrastructure enable biotech partners to accelerate drug discovery using secure, high-value machine learning models trained on decades of research data.
Abstract
Eli Lilly has introduced TuneLab, an AI platform that provides biotech companies access to drug discovery models trained on more than $1 billion of proprietary research data. The platform uses federated learning to let partners apply and improve these models without exchanging raw datasets. At the same time, Lilly is building one of the most powerful AI supercomputers in the pharmaceutical industry to support large-scale model development. Together, these moves represent a shift toward AI-augmented scientific workflows, where data, compute, and collaboration accelerate early-stage drug discovery.
Key Points
- TuneLab provides access to high-value drug discovery models trained on decades of internal R&D data.
- Federated learning protects proprietary data, allowing partners to use and strengthen models without sharing raw files.
- A large-scale NVIDIA-powered supercomputer supports model training, simulation, and scientific AI agents.
- The initiative lowers barriers for early-stage biotechs, offering capabilities that previously required significant infrastructure.
- The broader shift is toward AI-augmented research, where discovery speed depends on both scientific insight and computational scale.
Introduction
Artificial intelligence has been described as a breakthrough for drug discovery for years, but the impact has often been limited by a practical constraint: most AI models do not have access to the kind of large, high-quality experimental datasets required to make accurate predictions in real biological systems. Creating those datasets takes decades of research and significant investment. Very few organizations have them.
Eli Lilly has begun to change this dynamic. Through Lilly TuneLab, the company is making AI models trained on more than $1 billion worth of proprietary drug discovery data available to selected biotech partners. The models are trained on internal drug disposition, safety, and preclinical research data spanning hundreds of thousands of molecules. In parallel, Lilly is building what is positioned to be the most powerful AI supercomputer owned by a pharmaceutical company, using NVIDIA’s DGX B300 systems to support model development at industrial scale .
This combination changes what's possible in early-stage drug research.
The Core Problem
Drug discovery involves a long sequence of experiments that determine whether a molecule has the right biological and chemical characteristics to become a safe and effective medicine. Predicting outcomes like absorption, distribution, and toxicity has traditionally relied on iterative lab testing. This takes time and large teams, and the feedback loop can stretch out over many years.
AI can shorten these cycles, but only if the models are trained on dense, coherent, real experimental data. Most early-stage biotechs do not have that data. As a result, AI has often looked promising in theory but difficult to apply reliably in practice.
Lilly’s Approach: TuneLab
TuneLab provides access to Lilly’s AI models in a secure environment. Rather than sharing raw data, the platform uses federated learning, a framework where:
- Lilly’s global model is distributed to each partner
- The model trains on the partner’s data locally
- Only model updates are shared back to the central system
- No proprietary datasets are exchanged
This allows smaller companies to leverage high-value models while keeping their own data private, and it improves the models for everyone participating over time.
Early partners include Circle Pharma, which is applying TuneLab to design cell-permeable macrocycle therapies for difficult-to-treat cancers , and insitro, which is collaborating with Lilly to develop machine learning models that can predict the in vivo properties of small molecules.
This creates a continuous improvement cycle:
Each company contributes to a shared model ecosystem without sharing its proprietary data.
Together, TuneLab and Lilly’s AI supercomputing infrastructure create a two-layer system for accelerated research: shared, continuously improving models at the collaboration layer, and large-scale training capacity at the compute layer.
The Supercomputer and Why It Matters
Lilly is also constructing a large-scale AI supercomputer powered by more than 1,000 NVIDIA GPUs. It supports training biomedical foundation models, molecular property predictors, simulation workflows, and agentic AI systems that assist in experiment planning and design.
Several design choices stand out:
| Design Choice | Why It Matters |
| On-premise, not cloud-hosted | Treats compute as a strategic asset rather than a rented capability |
| Federated architecture extension | Aligns compute capabilities with TuneLab collaboration workflows |
| Runs on renewable energy | Reduces energy cost volatility and supports sustainability commitments |
This signals a shift: the boundary between pharmaceutical science and advanced computing is narrowing.
Why This Matters for Data and Analytics Work
This initiative is not only about drug development. It demonstrates several broader trends that are increasingly relevant across data-driven fields:
- Model access can substitute for dataset access.
High-value models can be shared without sharing raw data. - Federated learning is becoming operational reality.
Not a research idea, but production infrastructure. - Owning compute capacity can be a strategic differentiator.
Especially in fields where GPU access is scarce and timelines matter. - AI is moving from “tool” to “collaborator.”
As Lilly notes, the goal is to enable scientists to reason, plan, and test at larger scale, not replace them.
This is a shift toward AI-augmented scientific thinking.
Final Thoughts
Drug discovery remains one of the most complex problem domains in science. The challenge has never been lack of hypotheses. It has been the time, cost, and scale required to test those hypotheses in ways that are reliable and repeatable.
By pairing proprietary datasets with federated learning and large-scale compute infrastructure, Eli Lilly is moving toward a model where scientific progress is accelerated not just by new experiments, but by learning from every experiment ever run.
This may not eliminate the complexity of drug development. But it changes the pace at which meaningful work can be done.
The future of medicine may rely not only on discovering new treatments, but on discovering new ways to scale the process of discovery.
Link
Sources
Our new supercomputer could change the future of medicine
Circle Pharma Announces Agreement with Lilly to Further Enhance AI/ML Capabilities and Accelerate Development of New Oral Macrocycle Therapies
insitro partners with Lilly to build first-in-kind machine learning models to advance small molecule drug discovery
Lilly Deploys World’s Largest, Most Powerful AI Factory for Drug Discovery Using NVIDIA Blackwell-Based DGX SuperPOD
Further Reading
Eli Lilly and Nvidia team up to build 'most powerful' supercomputer in pharma
Eli Lilly invites biotechs to use its new AI platform to help develop their own drugs
Eli Lilly Offers Access to AI Models Trained on $1B Worth of Proprietary Drug Discovery Data
Eli Lilly Partners with insitro on Machine Learning Models for Small Molecule Discovery
Lilly opens up AI-trained drug-discovery models to biotechs for free
Lilly to give biotech startups access to AI tools