Eli Lilly's Fuchs Urges 1,000-GPU Supercomputing Push for Pharma R&D

2 articles · Updated · Emerj Artificial Intelligence Research · Aug 17

Thomas Fuchs said pharma must treat AI compute as core scientific infrastructure, arguing legacy IT cannot support frontier-scale models needed to speed discovery and improve manufacturing.
More than 95% of drug programs fail and a successful medicine can cost over $1 billion, he said, while unpublished failed experiments hold critical training data that current systems largely cannot exploit.
Fuchs drew a line between tools: LLMs fit documentation and regulatory work, while molecular, diffusion and physics-based models are needed for actual drug design and prediction.
1,000 B300 GPUs at Lilly remove limits on model size and simulation depth, he said, while AI-guided manufacturing has already delivered millions of additional doses through process optimization.
The broader gap is industry-wide: NSF this month committed $100 million to regional AI hubs, and FDA has reviewed more than 500 AI-related submissions since 2016, raising infrastructure and reproducibility demands.