Roche Builds AI Factory With NVIDIA to Accelerate Drug Development
Roche Holding AG is deploying 2,176 high-performance GPUs across U.S. and European facilities, bringing its total AI computing capacity to more than 3,500 NVIDIA Blackwell GPUs. The Swiss pharmaceutical company said Monday the infrastructure will speed drug discovery, improve manufacturing efficiency, and enhance diagnostic capabilities.
The scale marks one of the largest AI deployments disclosed by a pharmaceutical company. Roche's investment builds on a strategic partnership with NVIDIA that began in 2023.
Where Speed Matters Most
Wafaa Mamilli, Roche's Chief Digital and Technology Officer, said faster development cycles directly translate to patients receiving treatments sooner. "In healthcare, time is the most critical variable," Mamilli said.
The AI factory combines computing power with Roche's existing scientific expertise to streamline research, manufacturing, and diagnostics across the organization.
Lab-in-the-Loop: Testing at Scale
Roche is implementing a "Lab-in-the-Loop" approach where AI models work alongside real-world biology and chemistry experiments. Using NVIDIA's BioNeMo platform, scientists can test hypotheses at scale and compress discovery timelines.
In manufacturing, the company is deploying digital twins - virtual replicas of production systems - to optimize factory design and operations. The infrastructure also powers digital pathology systems that analyze large datasets to identify disease patterns.
Aviv Regev, head of Genentech Research and Early Development, said the expanded computing capacity will enable more advanced predictive models. Scaling the Lab-in-the-Loop strategy could significantly shorten the path from biological insights to treatments, Regev added.
Cancer Detection in 24 Hours
Droplet Biosciences, a diagnostics company, announced a separate collaboration with NVIDIA to detect residual cancer within 24 hours of surgery. The approach uses lymphatic fluid collected immediately after tumor removal, rather than blood plasma tested weeks later.
This timeline compression represents a practical application of AI infrastructure in clinical settings. Traditional minimal residual disease tests require weeks to deliver results.
Learn more about AI for Healthcare and AI for Science & Research.
Your membership also unlocks:
