Bristol Myers Squibb Just Gave Every One of Its Scientists Access to the Biggest AI Computer in Drug Research

Key points

• Bristol Myers Squibb (BMS) is adding a second NVIDIA DGX SuperPOD, built on eight Vera Rubin NVL72 rack-scale systems, to create what NVIDIA calls the most powerful AI cluster in life sciences.
• The new system delivers up to ten times the computing output per megawatt compared to the hardware it replaces.
• Every scientist at BMS will get access, with no waiting lists and no usage caps.
• BMS has run its first AI supercomputer for about three years, cutting weeks off drug-target research and expanding its library of cancer-fighting molecules.
• Researchers will be able to kick off complex predictions by typing plain-English requests rather than needing deep computing expertise.

Erin Davis has a new nickname for Bristol Myers Squibb's computing setup: the "SuperDuperPOD." It sounds playful. The science behind it is not.

BMS, one of the world's largest pharmaceutical companies, already runs a major AI computing cluster. This week the company announced it is adding a second, bigger one on top. Built on eight DGX Vera Rubin NVL72 units, rack-sized computers packed with Rubin GPUs (the specialised chips that do AI's heavy number-crunching) and Vera CPUs (general-purpose processors directing the overall workload), the new system delivers up to ten times the computing output per megawatt compared to what it replaces. Our 21 July story on NVIDIA's Vera Rubin architecture explained why that energy efficiency matters as AI workloads balloon.

That headline number matters in drug discovery because some AI predictions run across millions of possible molecules simultaneously. But the part Davis is most excited about is not raw power. It is access.

"Instead of equipping a small group of researchers with access to the supercomputer, we're opening it up to literally every scientist," says Davis, who leads research business insights and technology at BMS. "No one has to wait, and no one is told they have a limit."

Previously, site-specific restrictions left over from old acquisitions meant many researchers could not easily reach the system. The new setup connects BMS labs globally into a single platform. A team in Lawrenceville, New Jersey, and colleagues in San Diego, California, will draw from the same pool of data and models. Researchers can initiate complex predictions by typing a question in plain English.

What does this mean for patients waiting on new medicines?

Faster access to computing should shorten the gap between scientific ideas and testable drug candidates. Over three years with its first AI cluster, BMS cut weeks of manual target-identification work and built out a library of CELMoD compounds, molecules designed to break down cancer-causing proteins, with applications in blood cancer treatment. Payal Sheth, who took on the role of senior vice president of therapeutic discovery sciences in January, describes a "Predict First" approach: scientists model a molecule's likely behaviour before committing lab time to building it. Dead ends get weeded out early, keeping expensive experiments focused on the most promising candidates.

BMS is also building what it calls agentic workflows, where AI agents (software that can carry out multi-step tasks, searching data and surfacing insights across departments) connect learning from separate research programmes. Every experiment, Sheth says, now feeds a shared intelligence system that compounds over time. That cumulative loop, she notes, did not exist at the start of her career.

Davis's motivation is personal as well as professional. Her father died of Alzheimer's disease five years ago. BMS has a significant investment in brain health research, a notoriously hard area. The compute, she says, is not the goal. It is the means to get somewhere faster.

The combined system launches with a workload plan already mapped across small-molecule design, large-molecule design, clinical applications and biological simulations called digital twins. When BMS's chief digital officer asked Davis whether she could even saturate the new cluster, her answer was direct: "Just give us time."

"We didn't just buy this to have the biggest compute," Davis says. "Welcome to Limitless Compute."

The plain truth is that compute alone does not cure diseases. What BMS is betting on is that removing every bottleneck between a scientist's question and an answer worth testing will let the good ideas surface faster. That is a reasonable bet, and worth watching closely.