Bristol Myers claims biggest pharma AI cluster on Vera Rubin

Something Bristol Myers Squibb told STAT News this week is worth reading twice: it is the third pharma company in nine months, after Eli Lilly and Roche, to claim it is building the industry's largest Nvidia AI supercomputer. Whatever else this is, it is an arms-race announcement, and the framing deserves the hedge STAT gave it.

The kit itself is real. BMS is deploying its second Nvidia DGX SuperPOD, this one built on eight DGX Vera Rubin NVL72 systems, which the company says delivers up to ten times greater performance per megawatt than the infrastructure it replaces. The workloads BMS is aiming at span oncology, hematology, cardiovascular, immunology and neuroscience. According to a post on Nvidia's blog, the direction of travel is away from stand-alone tools like protein structure prediction and toward foundation models the company can aim at multiple therapeutic areas.

Greg Meyers, chief digital and technology officer at BMS, framed the expansion in practical terms. After nearly three years running the earlier SuperPOD, he said the company had 'consumed all the space we had.' That is the more revealing quote than the standard 'deliberate bet on AI' line, because it suggests real internal demand rather than a headline purchase. Erin Davis, VP of research business insights and technology, said the goal is to open the system up to 'literally every scientist' rather than a small internal elite.

The honest caveat is that a bigger cluster is not the same as a shipped medicine. Three years into the prior SuperPOD, the reporting does not give you a named candidate in the clinic that came out of it, nor does it disclose the deal's dollar figure, the total GPU count on the eight NVL72 racks, or how BMS's foundation-model program compares with the Lilly and Roche builds. Take the 'most powerful in life sciences' framing as marketing until the drug pipeline shows it.

The direction worth watching is who is not in the room. If the big pharma houses now own their own frontier-scale AI factories, the pressure on mid-tier drugmakers to rent equivalent capacity from a cloud provider, or to partner into someone else's cluster, gets a lot sharper over the next year.