BMS bets big on Nvidia's next-gen Vera Rubin chips to supercharge drug R&D

Bristol Myers Squibb is doubling down on its AI ambitions, becoming the first life sciences company to buy an Nvidia DGX SuperPOD built on the brand-new Vera Rubin architecture. The move isn’t just a hardware refresh; it’s a direct response to a compute crunch that has its existing, multi-year-old Nvidia cluster running at full capacity.

Erin Davis, BMS’s VP of research business insights, put it bluntly: the current infrastructure is maxed out, thanks to insatiable demand from large-scale molecular predictions and internal foundation model training. The new system, packing eight DGX Vera Rubin NVL72 racks, promises to blow the doors off those constraints. Chief Research Officer Robert Plenge framed the impact in practical terms: “Maybe before we could do 10 and now we can do dozens,” referring to the number of early-stage drug candidates his teams can evaluate.

This isn’t speculative tech hype. BMS is already using AI across virtually all its small-molecule programs and most large-molecule work, from target identification to lead optimization. The company claims AI-driven target ID has already shaved several weeks off manual research, and its “Predict First” methodology relies entirely on models to filter out dud molecules before anyone touches a pipette. Payal Sheth, SVP of therapeutic discovery, explained it’s about multi-parameter optimization, ensuring precious lab time is spent only on compounds with the highest probability of success.

The payoff isn’t theoretical. BMS credits its AI tools with helping discover an experimental sickle cell disease treatment that likely wouldn’t exist otherwise. They’ve also slashed the time to produce clinical trial medicines by 20-30%, with a target of 50% in the coming years. By scaling access beyond a small cadre of computational researchers and removing frustrating wait times, BMS is making a clear bet that raw computing power is now a direct competitive advantage in the race to develop new drugs.