NOAA replaces HPE Cray supercomputers with Google Cloud H4D VMs for weather prediction.

The U.S. National Oceanic and Atmospheric Administration has selected Google Cloud to replace its on-premise supercomputing infrastructure for weather forecasting operations. NOAA claims it will be the first national weather prediction center to operate on commercial cloud infrastructure, though the UK's Met Office is pursuing a parallel migration to Microsoft Azure with a hybrid configuration. This shift marks a significant departure from the traditional reliance on government-owned or government-funded supercomputer systems that have historically powered weather forecasting services.

General Dynamics previously managed NOAA's weather prediction systems, which most recently consisted of HPE Cray supercomputers deployed across data centers in Virginia and Arizona. These machines—named Dogwood and Cactus—delivered nearly 14 petaflops of computational capacity for weather prediction operations.

NOAA plans to transition its Weather and Climate Operational Supercomputing System, operated by the National Weather Service division, to Google Cloud by December 2027. The migration will include all associated software currently used to generate weather data for analysis by the NWS.

The agency anticipates that cloud-based infrastructure will accelerate operational forecasting by removing traditional bottlenecks inherent to on-premise systems. In a statement, NOAA Administrator Neil Jacobs explained that "cloud-based high-performance computing will accelerate the transition of research into operations by eliminating traditional bottlenecks of on-premise systems." The cloud architecture will also provide operational flexibility, allowing NOAA to scale compute capacity upward during tropical storm season and reduce it during calmer periods.

As part of the contract terms, NOAA gains access to Google DeepMind's AI tools to develop its first AI-driven weather forecasting system, the AI Global Forecast System. This system is designed to deliver accurate weather forecasts using 99.7% fewer computational cycles than traditional models and generate forecasts in minutes rather than hours.

The National Weather Service has already begun upgrading downstream software components to operate in cloud environments. In March, NWS awarded contracts to Accenture and Booz Allen Hamilton to develop cloud-native software platforms—HIVE and CIRRUS—for field offices to analyze data and issue alerts, replacing the existing on-premise systems.

Google will deploy Google Cloud H4D virtual machines based on AMD Epyc processors to support NOAA's workloads. These instances run under a hypervisor integrating Google's networking and orchestration capabilities, enabling synchronization for large-scale jobs comparable to traditional supercomputer operations. Google offers H4D access at pricing as low as 3 cents per core-hour without long-term commitments. For supercomputing workloads, customers can leverage Cluster Toolkit for cluster deployment and Cluster Director for ongoing management, while Google Cloud Batch handles job queuing, scheduling, and resource provisioning.