Bristol Myers Squibb is deploying a second NVIDIA DGX SuperPOD to expand the use of artificial intelligence across its global drug discovery organization. The new system will be built on eight NVIDIA DGX Vera Rubin NVL72 systems and integrated with the pharmaceutical company’s existing AI infrastructure.
The deployment is intended to give scientists across Bristol Myers Squibb access to a unified AI platform for running predictions, training proprietary models and building agentic research workflows. The company plans to make the computing environment available across research sites rather than restricting access to a small group of computational specialists.
Each rack-scale system combines NVIDIA Vera central processing units with Rubin graphics processing units. NVIDIA said the configuration can deliver up to 10 times the performance per megawatt of the infrastructure it replaces.
Bristol Myers Squibb will also use NVIDIA BioNeMo Agent Toolkit, a software platform designed for biological and pharmaceutical AI applications. The combination of computing infrastructure and life sciences software will support research across the full drug discovery pipeline.
The new platform is expected to help scientists evaluate larger chemical spaces, run more complex predictions and shorten research cycles. Bristol Myers Squibb wants researchers to focus more of their time on scientific decisions rather than managing computing capacity and technical infrastructure.
The company has operated its first NVIDIA DGX SuperPOD for approximately three years. Bristol Myers Squibb said the system has already helped researchers reduce the time required for AI-enabled target identification from weeks of manual work.
Researchers have also used AI to expand the company’s library of CELMoD compounds. These molecules are engineered to selectively degrade disease-causing proteins and are being investigated across blood cancers and other diseases.
Bristol Myers Squibb applies AI during lead optimization through an approach known internally as “Predict First.” Scientists use computational predictions to prioritize which molecules should be synthesized and tested in the laboratory.
The process evaluates multiple characteristics before experiments begin, helping teams exclude molecules that are less likely to meet the required profile. This can direct laboratory resources toward candidates with a higher probability of advancing through development.
The expansion is also intended to support Bristol Myers Squibb’s development of proprietary foundation models. The company said its existing infrastructure has reached capacity as teams run large-scale predictions involving complex molecules.
Bristol Myers Squibb plans to combine its current SuperPOD and the new Vera Rubin-powered system into a single environment accessible from its sites worldwide. A unified data layer will allow teams in different locations to use models and information generated by other research programs.
The company is seeking to remove barriers created by site-specific technology restrictions and the need for advanced computational expertise. NVIDIA Mission Control will manage the infrastructure, while AI-native interfaces are expected to let researchers initiate complex analyses using natural-language instructions.
The shared platform is also designed to create a cumulative learning system across Bristol Myers Squibb’s research organization. Data generated by one program or location can inform models and decisions used by scientists working on other diseases, drug candidates and therapeutic approaches.
Agentic AI workflows could further connect information across traditionally separate research functions. The company envisions specialized AI agents using Bristol Myers Squibb’s internal knowledge to assist scientists with research, analysis and decision-making across multiple programs.
Bristol Myers Squibb emphasized that AI will augment rather than replace human scientific judgment. Researchers will remain responsible for reviewing results, identifying limitations and determining how predictions should influence experiments and development decisions.
The company has mapped the new computing capacity across several research areas, including small molecules, large molecules, clinical applications and digital twins. This allocation is intended to ensure that the infrastructure supports practical projects throughout the research and development lifecycle.
The investment reflects growing demand for high-performance computing within pharmaceutical research. Drug developers are increasingly using AI to analyze biological information, design molecules, prioritize experiments and apply learnings across larger portfolios.
KEY QUOTES:
“Instead of equipping a small group of researchers with access to the supercomputer, we’re opening it up to literally every scientist. No one has to wait, and no one is told they have a limit. We’re saturated. We’re in production with some very large-scale predictions around large molecules. We’re building our own foundational models, and that takes a lot of GPUs. When you as a scientist can go to an army of well-vetted, fully trained virtual scientists that have BMS knowledge baked in, now you’re a whole team in and of yourself. You still have to have that human brain driving things, still looking for caveats and gotchas, still teaching them how to utilize knowledge. But this takes up the capabilities of individual humans substantially.”
Erin Davis, Vice President of Research Business Insights and Technology at Bristol Myers Squibb
“We use predictions as a way to prioritize synthesis of molecules with multi parameter optimization, to weed out molecules that wouldn’t necessarily meet the property landscape we’re working towards. This ensures precious laboratory experiments are aligned with progressing molecules that have the highest probability of success. The compute infrastructure is what connects all of our scientists together and ensures that our learnings are institutionalized. There’s a cumulative learning loop today in drug discovery that did not exist when I first started my career. Every project was treated differently, and there were discrete sets of learnings that did not compound into any kind of intelligence framework within discovery. Human instincts aren’t replaced, they’re augmented with more quantitative insights and predictions.”
Payal Sheth, Senior Vice President of Therapeutic Discovery Sciences at Bristol Myers Squibb
