Bristol Myers Squibb upgrades to Nvidia’s Vera Rubin AI architecture

The new infrastructure, which consists of eight DGX Vera Rubin NVL72 systems, marks a significant leap in processing power for the company. By combining Vera central processing units with Rubin graphics processing units, BMS aims to eliminate the capacity bottlenecks currently hindering its research teams. The hardware will be integrated with the company's existing three-year-old SuperPOD, creating a unified computing environment that allows researchers in locations like Lawrenceville, New Jersey, and San Diego to share data and model outputs seamlessly.

BMS executives emphasize that this investment is driven by the success of their "Predict First" methodology, which uses AI to filter out unsuitable molecules before they reach the laboratory. According to Chief Research Officer Robert Plenge, this computational approach has already reduced the time required to move candidates toward clinical trials by 20% to 30%, with potential for further gains. The system will also host Nvidia’s BioNeMo Agent Toolkit, providing specialized tools for molecular docking and protein-structure prediction.

Beyond raw performance, the transition addresses the rising costs of research operations. Greg Meyers, BMS’s chief digital and technology officer, noted that the Vera Rubin system offers up to 10 times the performance per megawatt compared to the company's older infrastructure. By reducing the specialized knowledge required to initiate complex tasks through natural-language instructions, BMS intends to democratize access to its computing resources, allowing scientists to focus on experimental outcomes rather than technical constraints.