Bristol Myers Squibb (BMS) has become the first life sciences company to deploy an NVIDIA DGX SuperPOD built on the next-generation Vera Rubin architecture. Made up of eight DGX Vera Rubin NVL72 systems, it delivers up to 10x the performance per megawatt of the infrastructure it replaces — and BMS plans to open it not to a select few but to "every scientist," extending AI across the full drug discovery pipeline. It builds on roughly three years of partnership since the company's first SuperPOD, and it underscores how competition is shifting from model benchmarks to the race for compute infrastructure.

Bristol Myers Squibb (BMS) announced on July 20 that it is building a new NVIDIA DGX SuperPOD based on the company's next-generation Vera Rubin architecture. The system comprises eight DGX Vera Rubin NVL72 units, which NVIDIA describes as "the most powerful and energy-efficient AI cluster in life sciences." BMS is the first life sciences company to adopt a Vera Rubin–based SuperPOD.

The Point Isn't a Bigger Model — It's Compute for Everyone

The most striking part of the announcement is less the hardware spec than the access model. Erin Davis, BMS vice president of research business insights and technology, calls the combined system the "SuperDuperPOD." "Instead of equipping a small group of researchers with access to the supercomputer, we're opening it up to literally every scientist," she said. "No one has to wait, and no one is told they have a limit."

Adopter Bristol Myers Squibb (BMS)
System 8x DGX Vera Rubin NVL72 (its 2nd DGX SuperPOD)
Performance Up to 10x per megawatt vs. replaced infrastructure
Partnership ~3 years since first DGX SuperPOD
Coverage Oncology, hematology, cardiovascular, immunology, neuroscience

What Changes — 'Predict First' and a Unified Data Plane

BMS has run its earlier DGX SuperPOD for about three years, with results to show for it. AI-enabled target identification already saves scientists weeks of manual work, freeing them to focus on higher-value scientific decisions. The team has used AI to expand its library of CELMoD compounds — molecules engineered to selectively degrade cancer-causing proteins — opening new targets and candidate medicines across blood cancers and beyond.

Payal Sheth, who expanded into the role of senior vice president of therapeutic discovery sciences in January, describes applying a "Predict First" methodology in lead optimization. Design-stage predictions gate experiments, weeding out molecules that don't meet the target property profile so that scarce lab work concentrates on the candidates with the highest probability of success.

The real lever is access, not raw compute. BMS is combining its existing SuperPOD and the new Vera Rubin system into a single data plane reachable from every BMS site globally. Managed through NVIDIA Mission Control, the environment lets researchers initiate complex predictions in plain English.

Why It Matters — The Center of Gravity Moves to Infrastructure

The deployment is another signal that AI competition is moving beyond benchmarks toward securing compute. Vera Rubin sharply raises performance per megawatt, letting BMS run larger, more sophisticated workloads without a proportional rise in power draw. Davis says large-molecule predictions and building the company's own foundation models have left existing resources "saturated," and she has already mapped an allocation plan for the new system — spanning small- and large-molecule design, clinical applications, and digital twins.

The stack includes the NVIDIA BioNeMo Agent Toolkit, supporting not just predictions and model training but agentic workflows across the drug discovery pipeline. "Agents don't care — they go all across," Davis said. "Now we can learn from decisions across the silos and across programs, and that is a huge game-changer."

Outlook

BMS frames the approach as "hybrid intelligence": computational systems handle data-intensive execution while human researchers retain direction, interpretation, and the judgment calls that require deeper expertise. "Human instincts aren't replaced," Sheth said, "they're augmented with more quantitative insights and predictions." With a drugmaker now emerging as a major buyer of NVIDIA's latest infrastructure — beyond cloud providers and AI labs — the spread of high-performance computing into scientific and industrial domains looks set to continue.

Related Reading · Official Sources
· NVIDIA Blog — BMS Building Life Science's Most Advanced AI Factory on Vera Rubin (official, Jul 20)
· NVIDIA — DGX Vera Rubin NVL72 product page
· NVIDIA — DGX SuperPOD overview
· Pharmaceutical Executive — BMS Expands NVIDIA Collaboration to Build AI Factory
  • BMS is the first life sciences company to deploy a Vera Rubin–based NVIDIA DGX SuperPOD
  • Eight DGX Vera Rubin NVL72 systems deliver up to 10x performance per megawatt vs. the infrastructure they replace
  • The old and new SuperPODs merge into a single data plane accessible from every BMS site worldwide
  • AI target identification already saves weeks; applied to CELMoD compounds and a "Predict First" methodology
  • BioNeMo Agent Toolkit powers agentic workflows across the drug discovery pipeline
  • Competition's center of gravity is shifting from model performance to securing compute infrastructure