🧠 Food for thought

Implications, context, and why it matters.

Roche’s AI push is more than just hardware

• Roche’s Genentech, its US biotech subsidiary, began a multi-year research collaboration with NVIDIA in 2023. The work aims to speed drug discovery and development by tuning Genentech’s models on NVIDIA DGX Cloud (NVIDIA’s cloud-based AI supercomputing service) and using NVIDIA BioNeMo (a platform for building AI models for biology and chemistry) for generative AI applications 1.
• The work uses a ’lab in a loop’ model where experimental data trains AI models. The models then produce predictions for lab testing, which keeps the cycle moving 1.
• The plan includes ’agentic AI’ (AI systems that can plan and act with minimal human direction) as an active collaborator. These systems can run workflows such as selecting clinical trial candidates or managing medication regimens without direct human prompting 2.

The AI factory model is moving from the cloud to the enterprise

• Roche is installing GPUs on premises (installing AI computing hardware in its own data centers rather than relying only on third-party cloud providers). This keeps sensitive data and intellectual property in-house.
• NVIDIA wants on-premises accelerated computing (using specialized chips like GPUs to speed up AI and other compute-heavy workloads) to become common in enterprise data centers. Its CEO said the approach is reshaping on-premises data-center design 3.
• Other organizations are using the same ’AI factory’ setup. Eli Lilly and the Mayo Clinic are building NVIDIA-based AI factories, including NVIDIA DGX SuperPOD systems, preconfigured AI data-center clusters from NVIDIA 4.
• Supporters expect this model to change drug-development costs. Failure rates run around 90%, so more predictable research could reduce wasted spending 1.