Bristol Myers Squibb adds NVIDIA Vera Rubin SuperPOD for AI-driven drug discovery

Bristol Myers Squibb is expanding its internal AI computing base with a second NVIDIA DGX SuperPOD, this time built on eight DGX Vera Rubin NVL72 systems. NVIDIA said the deployment is intended to create what it describes as the most advanced AI cluster in life sciences, giving BMS researchers a unified platform for model training, prediction workloads and agentic workflows across the drug discovery pipeline.

The announcement matters because pharmaceutical AI projects often run into a practical bottleneck: the science may be ready for larger models and broader screening, but researchers still have to compete for specialized compute. BMS already operates a DGX SuperPOD, and NVIDIA says the new system will be combined with the existing environment into a single data plane available across BMS sites globally.

What BMS plans to do with the system

According to NVIDIA, the Vera Rubin-based deployment will support workloads from small- and large-molecule design to clinical applications and digital twins. The company says each of the eight rack-scale systems combines NVIDIA Vera CPUs and Rubin GPUs, with the new setup delivering up to 10 times the performance per megawatt of the infrastructure it replaces.

BMS is also expected to use NVIDIA software including BioNeMo Agent Toolkit and Mission Control. The goal is to make advanced computational biology tools easier for scientists to access, including by allowing researchers to start complex predictions with natural-language prompts instead of relying only on deeply specialized infrastructure teams.

Why it is notable

Drug discovery remains a long and expensive process, and AI infrastructure announcements should not be read as immediate clinical breakthroughs. Still, BMS is presenting the expansion as a production-scale move rather than an experiment. NVIDIA said the company has already used AI-enabled target identification to save scientists weeks of manual work and to expand a library of CELMoD compounds, a class of molecules designed to degrade cancer-causing proteins.

The broader signal is that major pharmaceutical companies are treating AI compute as core research infrastructure. Instead of limiting large-scale predictive models to small specialist teams, BMS says it wants to make those capabilities available to scientists throughout its organization. If the system performs as described, it could shorten iteration cycles in early research, help prioritize laboratory experiments and make institutional data more reusable across programs.

The announcement also gives NVIDIA another high-profile example of Vera Rubin moving into domain-specific AI infrastructure. For health technology, the important question will be whether larger AI clusters translate into better research decisions, not just bigger model runs.

Sources

• blogs.nvidia.com

Cover photo by Chokniti Khongchum on Pexels, used under the Pexels License.