Source: STAT News — 2026-07-20
Summary
Bristol Myers Squibb is deploying a second NVIDIA DGX SuperPOD cluster — this one built on eight DGX Vera Rubin NVL72 systems — delivering up to 10x the performance per megawatt of the infrastructure it replaces, to support R&D across oncology, hematology, cardiovascular, immunology, and neuroscience. BMS is the third drugmaker in nine months (after Eli Lilly and Roche) to claim it's building the "largest" or "most powerful" AI supercomputer in life sciences, in a partnership with NVIDIA that dates back roughly three years. BMS's chief research officer says AI has already cut time-to-clinical-testing by 20-30%, with the new compute aimed at pushing that toward 50%.
Key Takeaways
- The hardware: eight NVIDIA DGX Vera Rubin NVL72 rack-scale systems (Vera CPUs paired with Rubin GPUs), added on top of BMS's existing DGX SuperPOD from its original ~3-year-old NVIDIA deployment — not a replacement, an expansion.
- Efficiency claim: up to 10x greater performance per megawatt versus the infrastructure being supplemented, meaning the same power budget buys roughly an order of magnitude more usable compute.
- Concrete before/after from BMS chief research officer Robert Plenge: AI has already reduced time to produce a candidate ready for clinical testing by 20-30%, and Plenge expects that to climb toward 50% over the next several years as this new compute comes online.
- Practical translation of the compute increase, per Plenge: the additional power lets BMS screen far more drug candidates simultaneously in early-stage discovery — "maybe before we could do 10 and now we can do dozens."
- BMS is the third pharma company in nine months to claim the "largest AI supercomputer in life sciences" title, after Eli Lilly (October 2025) and Roche (March 2026) — a signal that this specific superlative is becoming a recurring marketing beat in the industry rather than a one-off claim.
Reel Script
Hook (17s / 39 words) A drug company just said AI has already cut the time to get a compound into clinical testing by up to 30 percent. Not a projection — a number their chief research officer is citing today, before the new hardware they just announced even turns on.
Core Concept (55s / 128 words) Bristol Myers Squibb is adding a second NVIDIA supercomputer cluster, built from eight DGX Vera Rubin systems — basically racks that pair NVIDIA's new CPU with its new GPU, wired together to act like one giant machine. The headline spec is performance per megawatt: this generation does roughly ten times more useful computation for the same amount of electricity as what BMS was running before. Why that matters for drug discovery specifically — testing whether a molecule binds a target, predicting how a protein folds, simulating toxicity — is brute-force computation. More compute per watt means BMS can run more of those simulations in parallel without needing a proportionally bigger power bill, which is the actual bottleneck in scaling this kind of research infrastructure.
Hands-On (60s / 138 words) Here's the part worth writing down: BMS's chief research officer says AI is already cutting the time to get a drug candidate ready for clinical testing by 20 to 30 percent, and he expects that to climb toward 50 percent once this new compute is fully deployed. He also gave a concrete before-and-after on throughput — in early-stage screening, where before they could run maybe 10 drug candidates through analysis at once, they can now run dozens. That's not a vague "AI will accelerate research" claim, it's a specific multiple on parallel screening capacity. Worth noting: BMS is the third major drugmaker in nine months — after Eli Lilly and Roche — to announce it's building the "largest" AI supercomputer in life sciences, so treat the superlative itself skeptically even while the efficiency and timeline numbers are real.
Takeaway (24s / 55 words) Strip away the "largest in pharma" marketing framing and there's a real number underneath: 20 to 30 percent faster to clinical testing, today, with a credible path to 50 percent. If you work anywhere near computational biology or pharma infrastructure, DGX Vera Rubin's perf-per-watt jump is the spec worth tracking, not the superlative.