Scaling Scientific R&D with AI Supercomputing Infrastructure — with Thomas Fuchs of Eli Lilly

Eli Lilly’s new AI‑ready supercomputing platform, built on NVIDIA’s DGX SuperPod B300 architecture, marks a decisive shift from legacy IT to a strategic compute engine. With roughly a thousand GPU nodes—each as powerful as seven million 1989 Cray machines—the system delivers unprecedented parallel processing for foundation and frontier models across discovery, development, and manufacturing. This partnership with NVIDIA positions the pharma giant to leverage its 150‑year data legacy, turning decades of experimental results into high‑resolution AI insights.

The supercomputer fuels several high‑impact use cases. In drug discovery, larger models can ingest both positive and negative experimental data, dramatically expanding the searchable chemical space beyond what traditional methods achieve. By learning from millions of failed experiments, AI designs molecules with better safety and efficacy profiles, reducing reliance on animal testing. In manufacturing, digital twins of drying processes and equipment have already cut cycle times, enabling millions of additional doses to reach patients faster and delivering immediate financial returns. These initiatives are tracked against clear metrics, linking compute investment directly to ROI.

Lilly’s approach reflects a broader industry trend: AI infrastructure is becoming a core capability rather than a background service. Secure, collaborative platforms like TuneLab allow vetted models to be shared with biotech partners while safeguarding proprietary data. The emphasis on negative data, model validation, and human‑in‑the‑loop oversight addresses hallucination risks and paves the way for more reliable, deterministic AI tools. As pharmaceutical R&D accelerates, the combination of massive compute, robust data governance, and strategic partnerships will reshape timelines, costs, and ultimately patient outcomes.