Bristol Myers Squibb

Bristol Myers Squibb, a pharmaceutical giant specializing in innovative therapies (also present in Italy), has announced the development of its second NVIDIA DGX SuperPOD, an AI infrastructure aimed at expanding the use of artificial intelligence in all phases of new drug development. The new system will be built with eight DGX Vera Rubin NVL72, defined by the company as the most powerful and energy-efficient AI cluster in the life sciences sector.

Each rack-scale system integrates NVIDIA's Vera CPUs and Rubin GPUs, with a stated capacity up to 10 times greater per megawatt compared to the infrastructure it will replace. The SuperPOD will provide BMS scientists with a unified AI platform, with access to the NVIDIA BioNeMo Agent Toolkit for biological AI, useful for running simulations, training models, and managing agent workflows throughout the entire drug discovery pipeline.

According to Erin Davis, Vice President of Research Business Insights and Technology at BMS, the goal is not only to increase computing power, but to make it available to all the company’s researchers. "Instead of equipping a small group of researchers with access to the supercomputer, we are making it literally accessible to every scientist. No one has to wait, and no one is imposed a limit," Davis explained.

BMS has already been using a first DGX SuperPOD for about three years, which has produced tangible results in pharmaceutical research. AI has enabled the reduction of weeks of manual work in identifying therapeutic targets and has expanded the CELMoD compound library, molecules designed to selectively degrade proteins responsible for certain cancers, with applications in blood cancers and other diseases. The company also applies AI in the lead optimization stages through a methodology called Predict First. This approach utilizes computational predictions to select the molecules with the highest likelihood of success before synthesis and laboratory testing.

"We use predictions as a method to prioritize the synthesis of molecules with multi-parameter optimization to discard molecules that would not necessarily meet the properties we are trying to achieve. This ensures that valuable laboratory experiments are aligned with the developing molecules that have the highest probability of success," explained Payal Sheth, scientist and Senior Vice President of Therapy Discovery Sciences at BMS.

The new DGX SuperPOD will be integrated with the existing one into a single shared environment, with a single data plane accessible from all BMS sites worldwide. Previous limitations, such as site-specific restrictions or the need for advanced computational skills, will be replaced by AI-native tools managed through NVIDIA Mission Control.

"Today, in new drug discovery, there exists a cumulative learning cycle that was not there when I started my career," Sheth explained. "Each project was treated differently, and knowledge was acquired in a fragmented manner, without this combining into a shared framework within the discovery process."

Scientists will be able to initiate complex predictions using natural language. In this way, the data generated in one laboratory can feed models used by teams in other locations, creating a cumulative learning cycle that BMS considers central to accelerating research.

BMS plans to use the new system in various areas, from designing small and large molecules to clinical applications, up to digital twins. According to Davis, agent workflows will allow learning from decisions made in different research programs and sharing knowledge among teams that previously worked in a more isolated manner.

The company has already planned the allocation of computing power from the new cluster for various research modalities. "We didn't purchase this system just to have the largest cluster," Davis explained. "The SuperDuperPOD will be present at every node of the drug discovery process."