Bristol Myers Squibb Building Life Science Industry’s Most Advanced AI Factory on NVIDIA Vera Rubin

Erin Davis has a nickname for the thing Bristol Myers Squibb just announced, and it tells you everything about the ambition behind it. She calls it the SuperDuperPOD. BMS already runs one of the largest AI clusters in life sciences, and rather than resting on that, the company is adding a second NVIDIA DGX SuperPOD, this one built on eight DGX Vera Rubin NVL72 systems. The pitch from Davis, who is vice president of research business insights and technology, is not about raw horsepower for a chosen few. It's about erasing the wait entirely. "No one has to wait, and no one is told they have a limit," she says.

The hardware itself is not a modest upgrade. Eight rack-scale systems combining NVIDIA Vera CPUs and Rubin GPUs deliver up to 10 times the performance per megawatt of the infrastructure being replaced. That capacity feeds a unified AI platform, including the NVIDIA BioNeMo Agent Toolkit, that researchers can use for predictions, model training and agentic workflows spanning the entire drug discovery pipeline. The idea is to let scientists worry about chemistry and biology, not about scheduling time on a shared machine.

This is not a leap of faith. BMS has run a DGX SuperPOD for roughly three years, and the payoff has been concrete. AI-driven target identification already shaves weeks off manual work. The company has used AI to grow its library of CELMoD compounds, molecules designed to selectively degrade cancer-causing proteins, with uses in blood cancer and beyond. Payal Sheth, who became senior vice president of therapeutic discovery sciences in January after years inside BMS's drug discovery labs, describes a method called Predict First, where computational predictions determine which molecules are worth synthesizing at all, so lab time goes toward candidates with the best odds.

Davis's own path into this work is personal, not just technical. A computational chemist by training, she spent about 15 years on the vendor side at companies including ChemAxon, Schrödinger and X-Chem, watching pharma companies run into the same bottleneck everywhere she went: it was never the technology holding things back, it was getting that technology into scientists' hands. Her father died five years ago from Alzheimer's, and she talks about how even symptom relief along the way, short of a cure, would have eased suffering for the whole family. BMS's investment in brain health, a famously difficult research area, is part of what she's building toward.

What makes the new system more than just bigger compute is how it's being stitched together. Davis's team is merging the existing SuperPOD and the new Vera Rubin system into one environment, a single data plane reachable from every BMS site worldwide. That matters because older barriers, leftover site restrictions from past acquisitions, the need for deep computational expertise, are being replaced with AI-native tools managed through NVIDIA Mission Control, letting researchers run complex predictions in plain English. Sheth calls it a cumulative learning loop that simply didn't exist earlier in her career, when every project's lessons stayed siloed. Now a dataset from a program in Lawrenceville, New Jersey can inform work happening in San Diego. Agentic workflows push that further, since agents can cross silos and programs in ways individual scientists structurally couldn't before.

Davis insists the buildout isn't about bragging rights over compute size. There's a detailed allocation mapped across small and large molecule design, clinical applications and digital twins, node by node. When BMS's chief digital and technology officer, Greg Meyers, asked whether she could actually saturate a system this large, her response left no room for doubt: "Just give us time."