Bristol Myers Squibb (BMS) has announced the acquisition of an Nvidia DGX SuperPOD based on the Vera Rubin architecture, the latest generation of artificial intelligence infrastructure from the chipmaker. This investment, the first of its kind in the life sciences sector, will enable BMS to integrate advanced AI capabilities across its entire drug discovery and development process, from target identification to candidate optimization and drug manufacturing.
The new cluster consists of eight DGX Vera Rubin NVL72 systems, each combining Nvidia Vera central processing units (CPUs) with Rubin graphics processing units (GPUs). This configuration delivers up to ten times the performance per megawatt compared to BMS's previous infrastructure, which already included an earlier-generation SuperPOD. The company plans to operate both systems in a shared computing environment accessible from its research sites worldwide, eliminating the waiting times and access limits that characterized the previous system.
Greg Meyers, BMS's chief digital and technology officer, noted that computing needs have grown exponentially as the company deploys larger AI models across its research organization. Erin Davis, vice president of research business insights and technology, confirmed that current infrastructure is operating at full capacity, driven by large-scale predictions involving large molecules and the development of internal foundation models. With the new system, BMS aims to democratize access to AI: it will no longer be restricted to a small group of computational researchers but will be available to the entire research team without capacity constraints.
AI application in drug discovery is a strategic pillar for BMS. The company states that AI informs the design of all its small-molecule programs and most of its large-molecule programs. Artificial intelligence technologies are used for target identification, lead optimization, large-molecule predictions, and internal model development. For instance, AI-enabled target identification has reduced manual research work by several weeks. Robert Plenge, BMS's chief research officer, explained that the new system will allow evaluation of more drug candidates in early stages: 'Before we could do 10; now we can do dozens.'
This approach, called 'Predict First,' uses model-generated predictions to exclude molecules that do not meet required properties before candidates are selected for synthesis and laboratory testing. Payal Sheth, senior vice president of therapeutic discovery sciences, highlighted that predictions enable prioritization of molecule synthesis with multi-parameter optimization, ensuring that laboratory experiments focus on molecules with the highest probability of success. This method has reduced the number of compounds entering experimental testing, allowing researchers to concentrate on those with the greatest potential.
Additionally, BMS has used AI to expand its library of CELMoD compounds, designed to selectively degrade cancer-causing proteins. Computational modeling has helped examine a larger number of protein targets and potential compounds before deciding which candidates to pursue experimentally. The company also uses AI tools to shorten the time required to produce medicines for clinical trials, reducing it by 20% to 30%, with a goal of reaching 50% in the coming years. Plenge mentioned an experimental sickle cell disease treatment as an example of a drug that likely would not have been discovered without the company's AI tools.
The new Vera Rubin system will also provide access to Nvidia's BioNeMo Agent toolkit for biological and drug-discovery applications. BioNeMo includes capabilities for protein structure prediction, molecular generation, molecular docking, sequence analysis, and genomics. Researchers will be able to connect multiple computational tools within a single workflow. BMS has also introduced tools that allow prediction requests to be initiated using natural language instructions, reducing the need for specialized knowledge for complex computational tasks.
The unified infrastructure will allow data and model outputs generated at one site (for example, Lawrenceville, New Jersey) to be incorporated into models used by researchers in San Diego. This shared environment retains information from experiments and research programs across the organization. Sheth stated: 'The compute infrastructure is what connects all our scientists together and ensures that our learnings are institutionalized.' The two SuperPODs will operate through a common data environment, which will include information from experiments, clinical readouts, and research partnerships.
Adopting such advanced AI systems poses significant challenges in terms of technology integration, cybersecurity, and data management. Pharmaceutical companies need custom software platforms to orchestrate workloads, manage large volumes of data, and ensure the security of sensitive information. In this context, companies like Q2BSTUDIO, specialized in software development and technology, offer key services for implementing custom artificial intelligence solutions, as well as cloud infrastructure on AWS and Azure, and business intelligence tools like Power BI to visualize and analyze predictive model outputs. Cloud AWS/Azure provides the scalability and flexibility needed to train complex models, while cybersecurity is essential to protect research data and intellectual property. AI agents, meanwhile, can automate complex workflows, from predicting molecular properties to prioritizing experiments.
BMS's investment in Nvidia's Vera Rubin architecture marks a milestone in the application of artificial intelligence to life sciences. It not only demonstrates the growing appetite for high-performance computing in the pharmaceutical sector but also underscores the importance of having technology partners capable of effectively integrating these solutions. As more companies adopt similar approaches, collaboration between pharmaceutical firms, chip manufacturers, and software development companies will be crucial to accelerating the discovery of new treatments and improving research process efficiency.
The Vera Rubin system has no specific deployment date announced yet, but it is expected to become operational in the coming months, hosted in BMS or Nvidia data centers. The company plans to allocate the new computing capacity to small- and large-molecule design, clinical research, and digital twin applications, although details on the latter have not been provided. With this move, BMS positions itself at the forefront of digital transformation in the pharmaceutical industry, where artificial intelligence is no longer an option but a competitive necessity.
