The US National Oceanic and Atmospheric Administration (NOAA) has selected Google Cloud to run the infrastructure behind its weather forecasting operations, moving away from government-owned supercomputers. NOAA says it will become the first national weather prediction centre to operate on a commercial cloud platform, a shift it hopes will remove the bottlenecks of on-premise systems and allow computing capacity to scale up during events such as tropical storm season. The move follows a broader trend of national weather agencies exploring cloud infrastructure, with the UK's Met Office also transitioning to a hybrid setup on Microsoft Azure.

The contract replaces General Dynamics' management of NOAA's HPE Cray supercomputers, known as Dogwood and Cactus, which delivered nearly 14 petaflops of processing power from data centres in Virginia and Arizona. NOAA plans to complete the migration of its Weather and Climate Operational Supercomputing System, run by the National Weather Service, by December 2027, using Google Cloud H4D virtual machines built on AMD Epyc processors. The deal also gives NOAA access to Google DeepMind's AI tools to develop an AI-driven forecasting system claimed to use 99.7% fewer computing cycles, while separate contracts with Accenture and Booz Allen Hamilton are already updating downstream field-office software for the cloud.