{"id":4627,"date":"2026-07-20T13:46:21","date_gmt":"2026-07-20T13:46:21","guid":{"rendered":"https:\/\/salarydistribution.com\/machine-learning\/2026\/07\/20\/bristol-myers-squibb-building-life-science-industrys-most-advanced-ai-factory-on-nvidia-vera-rubin\/"},"modified":"2026-07-20T13:46:21","modified_gmt":"2026-07-20T13:46:21","slug":"bristol-myers-squibb-building-life-science-industrys-most-advanced-ai-factory-on-nvidia-vera-rubin","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2026\/07\/20\/bristol-myers-squibb-building-life-science-industrys-most-advanced-ai-factory-on-nvidia-vera-rubin\/","title":{"rendered":"Bristol Myers Squibb Building Life Science Industry\u2019s Most Advanced AI Factory on NVIDIA Vera Rubin\u00a0"},"content":{"rendered":"
\n

Erin Davis calls it the \u201cSuperDuperPOD.\u201d That\u2019s two things in one name: pharmaceutical giant Bristol Myers Squibb (BMS) already runs one of the largest AI clusters in life sciences, with serious results to show for it. And they\u2019re doubling down.<\/span><\/p>\n

BMS announced today it is deploying its second <\/span>NVIDIA DGX SuperPOD<\/span><\/a>, this one built on eight <\/span>DGX Vera Rubin NVL72 systems<\/span><\/a> \u2014 the most powerful and energy-efficient AI cluster in life sciences.<\/span><\/p>\n

\u201cInstead of equipping a small group of researchers with access to the\u00a0 supercomputer, we\u2019re opening it up to literally every scientist,\u201d says Davis, vice president of research business insights and technology at BMS. \u201cNo one has to wait, and no one is told they have a limit.\u201d<\/span><\/p>\n

The eight rack-scale systems, <\/span>each comprising NVIDIA Vera CPUs and <\/span>Rubin GPUs, deliver up to 10x the performance per megawatt of the infrastructure it replaces. It will give researchers at the global pharmaceutical giant access to a unified AI platform \u2014 including <\/span>NVIDIA BioNeMo Agent Toolkit <\/span><\/a>for biological AI \u2014 for running predictions, training models and powering agentic workflows across the full drug discovery pipeline.<\/span><\/p>\n

What Davis and other top BMS researchers are really after is what that access makes possible: faster cycles, bigger chemical spaces and a full drug discovery pipeline where researchers think about the science, not the logistics of lining up resources.<\/span><\/p>\n

The mandate, says Payal Sheth \u2014 a scientist who spent her career inside drug discovery labs before taking on an expanded role in January as senior vice president of therapeutic discovery sciences at BMS \u2014 is moving from \u201csort of this abstract position of what AI can do to actually translating that to measurable impact.\u201d<\/span><\/p>\n

BMS <\/span>has operated a DGX SuperPOD<\/span><\/a> for about three years, producing meaningful results. AI-enabled target identification already saves scientists weeks of manual work, freeing time to focus on the highest-value scientific decisions. BMS\u2019s team has used AI to expand its library of CELMoD compounds \u2014 molecules engineered to selectively degrade cancer-causing proteins, with applications in blood cancer treatment and beyond. This has opened the door to new targets and new potential medicines across a wider range of diseases. AI is also applied in lead optimization stages of drug discovery using a methodology Sheth calls \u201cPredict First,\u201d which informs experimental gating based on design predictions.<\/span><\/p>\n

\"\"
Teams at Bristol Myers Squibb review a computational model of a molecule\u2019s structure, part of the predictive design process that helps scientists anticipate how a molecule will behave in clinic. Image credit: Bristol Myers Squibb<\/figcaption><\/figure>\n

\u201cWe use predictions as a way to prioritize synthesis of molecules with multi parameter optimization,\u201d she explains, \u201cto weed out molecules that wouldn\u2019t necessarily meet the property landscape we\u2019re working towards. This ensures precious laboratory experiments are aligned with progressing molecules that have the highest probability of success.<\/span><\/p>\n

These research AI applications have significant impact on compute needs across the research organization.\u201cWe\u2019re saturated,\u201d Davis says. \u201cWe\u2019re in production with some very large-scale predictions around large molecules. We\u2019re building our own foundational models, and that takes a lot of GPUs.\u201d<\/span><\/p>\n

With the new system coming, Davis already has her pitch for researchers thinking about where to do their best work: \u201cWelcome to Limitless Compute.\u201d<\/span>
<\/span><\/p>\n

A computational chemist by training, Davis spent years doing the science before concluding the technology wasn\u2019t keeping up \u2014 and that she\u2019d rather go fix it. She spent roughly 15 years on the vendor side, building enterprise platforms at ChemAxon, Schr\u00f6dinger and X-Chem. At every company, pharma was wrestling with the same bottleneck.<\/span><\/p>\n

\u201cIt\u2019s not the technology,\u201d she says. \u201cThe challenge is how to get that into the hands of actual scientists and learn from it.\u201d<\/span><\/p>\n

She knows what\u2019s at stake personally. Her father died five years ago, she says, \u201ca very horrible death of Alzheimer\u2019s.\u201d BMS has a significant investment in brain health \u2014 a notoriously hard area. \u201cEven if he was still going to die,\u201d Davis says, \u201cif there was symptom remediation along the way, it would have saved suffering for everybody in the family. Dementia is especially cruel.\u201d<\/span><\/p>\n

Davis\u2019s team is combining the existing DGX SuperPOD and the new DGX Vera Rubin NVL72-powered system into a unified environment \u2014 a single data plane, accessible from every BMS site globally.\u00a0<\/span><\/p>\n

Barriers that made the earlier system hard to reach \u2014 site-specific restrictions left over from past acquisitions, the need for deep computational expertise \u2014 are being replaced with AI-native tooling<\/span> managed through <\/span>NVIDIA Mission Control<\/span><\/a>. Researchers will be able to initiate complex predictions in plain English.<\/span><\/p>\n

\u201cThe compute infrastructure is what connects all of our scientists together and ensures that our learnings are institutionalized,\u201d Sheth explains. Datasets from a program run in Lawrenceville, New Jersey, feed models that a team in San Diego, California, can draw on. The learnings \u201ccan be applied in context of any program we work on.\u201d<\/span><\/p>\n

\u201cThere\u2019s a cumulative learning loop today in drug discovery that did not exist when I first started my career,\u201d Sheth explains. \u201cEvery project was treated differently, and there were discrete sets of learnings that did not compound into any kind of intelligence framework within discovery.\u201d<\/span><\/p>\n

Today, BMS is using AI to expand that learning loop into a discovery system where every experiment, clinical readout, and partnership compounds into higher-conviction scientific decisions, faster.\u00a0<\/span><\/p>\n

Agentic workflows can further enhance the architecture of R&D.<\/span><\/p>\n

\u201cAgents don\u2019t care,\u201d Davis says. \u201cThey go all across. And that is a huge game-changer because now we can learn from decisions across the silos and across programs.\u201d<\/span><\/p>\n

\u201cWhen you as a scientist can go to an army of well-vetted, fully trained virtual scientists that have BMS knowledge baked in now you\u2019re a whole team in and of yourself.\u201d<\/span><\/p>\n

Human instincts, Sheth says, aren\u2019t replaced, \u201cthey\u2019re augmented with more quantitative insights and predictions.\u201d The ability to scale that with compute, she says, \u201cis where the excitement of the impact of AI is going to be fully realized.\u201d<\/span><\/p>\n

\u201cYou still have to have that human brain driving things,\u201d Davis adds, \u201cstill looking for caveats and gotchas, still teaching them how to utilize knowledge. But this takes up the capabilities of individual humans substantially.\u201d<\/span><\/p>\n

Davis says the new system has a plan already mapped to it: a detailed allocation across modalities, from small and large molecule design to clinical applications to digital twins. \u201cWe didn\u2019t just buy this to have the biggest compute,\u201d she says. \u201cThe SuperDuperPOD is basically at every node along the way.\u201d<\/span><\/p>\n

When BMS Chief Digital and Technology Officer Greg Meyers asked Davis whether she was sure she could even saturate the super-duper pod, her answer was direct.<\/span><\/p>\n

\u201cJust give us time,\u201d she told him. <\/span>
<\/span><\/p>\n

Featured image credit: Bristol Myers Squibb<\/span><\/em><\/p>\n<\/p><\/div>\n","protected":false},"excerpt":{"rendered":"

https:\/\/blogs.nvidia.com\/blog\/bristol-myers-squibb-building-life-science-industrys-most-advanced-ai-factory-on-nvidia-vera-rubin\/<\/p>\n","protected":false},"author":0,"featured_media":4628,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/4627"}],"collection":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/comments?post=4627"}],"version-history":[{"count":0,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/4627\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media\/4628"}],"wp:attachment":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media?parent=4627"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/categories?post=4627"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/tags?post=4627"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}