Overview
- Bristol Myers announced on Monday that it is acquiring a second NVIDIA DGX SuperPOD made from eight DGX Vera Rubin NVL72 systems and will combine it with its existing SuperPOD into a single, company‑wide platform.
- The company and NVIDIA say the Vera Rubin systems deliver up to 10 times more performance per megawatt than BMS’s prior setup, a claim the companies use to justify larger, more energy‑efficient model training for molecular design and imaging work.
- BMS will adopt NVIDIA’s BioNeMo platform and the BioNeMo Agent Toolkit to train proprietary foundation models and run agentic workflows that automate tasks such as target identification and validation.
- Executives point to measurable gains from the first SuperPOD—AI cut target‑identification work from weeks to days and reportedly reduced some compute costs by about 55%—and say the new purchase aims to widen researcher access so no team must wait for high‑end GPUs.
- Some outlets report the cluster will be colocated at Equinix and go online by January 2027 but BMS has not uniformly confirmed the hosting provider or that timeline, making those details provisional.