Roche's new NVIDIA deal locks in the industry's 'largest hybrid-cloud AI factory'

Roche said Monday it is expanding its partnership with NVIDIA to build what it calls the pharmaceutical industry's "largest hybrid-cloud AI factory," aimed at accelerating the development of new therapeutics and diagnostics.

The new pact boosts Roche's AI capacity with 2,176 Nvidia Blackwell graphics processing units (GPUs), which will be deployed across the United States and Europe. This brings the Swiss drugmaker's total GPU count to 3,000, which Roche says is the largest in the industry.

For drugmakers, the number of GPUs a company has determines how fast AI models can be trained, scaled and improved. Adding more GPUs leads to faster data processing and the ability to run more complex models.

Roche's new AI factory is a high-performance supercomputing platform designed to drive digital transformation across the organization, the pharma giant said in its press release. It aims to accelerate drug discovery, improve clinical trial operations and unlock data insights at scale to advance innovation and improve healthcare outcomes

The expanded partnership brings together NVIDIA's "world-class computing power" with Roche's scientific expertise to integrate AI across the value chain, from discovery to commercialization, Roche said.

Through the partnership, Roche will gain access to a range of NVIDIA technologies, including BioNeMo, NVIDIA's R&D platform. BioNeMo supports Roche's "Lab-in-the-Loop" approach, which allows researchers to test hypotheses at scale.

"By providing the massive computational power needed to continue to scale our Lab-in-the-Loop strategy -- a space we have pioneered for over five years -- our scientists can build more sophisticated predictive frontier models and further shorten the path from biological insight to life-saving medicine," Aviv Regev, Executive Vice President and Head of Genentech Research and Early Development, said in Roche's press release.

The pharma titan will also use NVIDIA's Omniverse digital twin technology to create virtual replicas of production lines, helping optimize factory design and processes. Roche's use of NVIDIA's Parabricks software will help analyze large datasets and identify subtle disease patterns in imaging and pathology workflows. Additionally, Roche will deploy NVIDIA's NeMo Guardrails to help maintain safety and reliability in its healthcare-focused conversational AI tools.

The new pact builds on a 2023 research collaboration between Roche's subsidiary, Genentech, and Nvidia to expand the use of GenAI to drive drug innovation. It also comes as several other Big Pharma players are turning to AI to accelerate R&D, reduce research costs and improve clinical success rates.

In January, pharma giant Eli Lilly teamed up with NVIDIA to build a $1 billion AI-powered drug discovery lab aimed at improving the traditionally slow, expensive drug discovery process. Less than two months later, Lilly launched LillyPod, the industry's most powerful supercomputer, powered by 1,016 Blackwell Ultra GPUs on NVIDIA's DGX SuperPOD, to advance scientific research.

Alivia Kaylor is a scientist and the senior site editor of Pharma Life Sciences.