Bristol Myers Squibb has announced the purchase of a new Nvidia AI supercomputer system called a DGX SuperPOD, built on Nvidia's Vera Rubin architecture, which was introduced at CES in January 2026. BMS will be the first life sciences company to own one. The price was not disclosed.
The system consists of eight connected rack units, each combining Nvidia's Vera processors with Rubin computing chips. Together, the eight units deliver roughly 28.8 exaflops of processing capacity, a number that is less important than what it means in practice: BMS can now run far more AI predictions, simultaneously, than it could before, without a proportional jump in electricity costs.
That last point matters. The new system delivers ten times more computing work per megawatt of electricity than the older SuperPOD it is joining. BMS's chief digital officer said it plainly: electricity is not getting cheaper, and this is essentially ten times more output for the same power bill.
The reason BMS needed the upgrade is straightforward. Its existing system was full. Scientists across its research sites were waiting for access to run calculations. The company now plans to connect both the old and new systems into a single shared computing environment, accessible by research teams across its sites in New Jersey, San Diego, and elsewhere.
What is BMS actually using AI for? The clearest example is what the company calls its "Predict First" approach. Before any molecule is physically made and tested in a laboratory, AI models run predictions to estimate whether the molecule has the right properties to be a useful drug. The ones that do not clear that bar are set aside. Only the candidates that look promising on paper proceed to physical synthesis and lab testing.
This approach now shapes every small-molecule drug programme at BMS and the majority of its large-molecule programmes. As recently as 2021, that figure was around five percent. The shift is not marginal.
The company's chief research officer said the new system will allow scientists to evaluate meaningfully more drug candidates in early development: "Maybe before we could do 10 and now we can do dozens." That is not a rounding error. In drug discovery, testing more early candidates cheaply is how you raise the odds that something actually works by the time it reaches clinical trials.
BMS also uses AI to find biological targets for drugs, meaning identifying the specific proteins or mechanisms that a drug should go after. This step, which traditionally required weeks of manual literature review and analysis, is now partially automated. The company says it has cut the time required to get from candidate identification to clinical trial-ready medicine by 20 to 30 percent, and it is targeting 50 percent reduction in the years ahead.
The wider industry context makes this purchase make more sense. Across pharma broadly, AI-originated drug programmes in clinical development have grown from roughly 24 programmes in late 2023 to over 173 by early 2026. One fully AI-designed drug, developed by Insilico Medicine for a lung disease, completed Phase IIa trials in February 2026 at a discovery cost of approximately $6 million. The traditional path to the same milestone typically costs $100 to $200 million and takes six to eight years.
BMS is not a small startup making big claims. It is one of the largest pharmaceutical companies in the world, and it is now treating AI computing capacity the same way it treats laboratory space: as core operating infrastructure that limits throughput when it runs short.
For anyone outside pharma, the relevant takeaway is the model itself. BMS is using AI to screen and filter before spending money on physical experiments. That logic, run predictions first and only proceed when the numbers look right, applies to a much wider range of industries than drug discovery. The specific tools differ, but the operating principle of using AI to reduce expensive physical or human work downstream is already being applied in manufacturing quality control, insurance underwriting, supply chain risk assessment, and materials procurement. BMS is simply further along than most.