Particle.news
Download on the App Store

Roche Launches Nvidia-Powered AI Factory With 2,176 New Blackwell GPUs, Claims Pharma’s Largest Footprint

Roche describes a production-grade system embedding accelerated computing across its value chain.

Overview

  • With the new on-premises deployment across the United States and Europe, Roche says its hybrid setup now exceeds 3,500 Blackwell GPUs.
  • The company details integrations that operationalize AI at scale, including NVIDIA BioNeMo for Lab-in-the-Loop research, Omniverse for manufacturing digital twins, Parabricks for diagnostics, and NeMo Guardrails for healthcare-grade conversational AI.
  • This expansion is presented as the next phase of a strategic NVIDIA collaboration that began in 2023, shifting from pilots to broad production use.
  • Roche declined to disclose total spending, while Forbes notes Jensen Huang’s $30,000–$40,000 per Blackwell estimate, implying chip costs of $65 million or more.
  • Roche cites internal results to validate the approach, including designing an oncology molecule 25% faster and using AI to predict immunogenicity for a therapy entering human trials.