For decades, NOAA sized its high-performance computing (HPC) systems using historical performance data from physics-based simulation workloads. The emergence of AI-driven forecasting disrupted this model, requiring NOAA to shift from backward-looking capacity planning to forward-looking projections amid uncertainty in AI timelines and requirements.

This session explores how NOAA prepared for that shift, enabling AI weather prediction in mid-2025 with the adoption of Google DeepMind’s GraphCast model. By placing an early bet on AI-ready infrastructure, NOAA was able to adapt quickly as AI workloads came online—without compromising its core HPC mission. The joint solution delivered by GDIT, VAST Data, and Supermicro also closed performance gaps in traditional HPC, demonstrating how a unified platform can support HPC and AI together.