The U.S. National Oceanic and Atmospheric Administration (NOAA) is phasing out its weather forecasting supercomputer and migrating to Google Cloud.
The National Oceanic and Atmospheric Administration (NOAA) has adopted Google Cloud as its primary provider of high-performance computing infrastructure for weather forecasting. The plan is to migrate from its previous operation, which relied on dedicated supercomputers, to a public cloud that allows for scaling computing resources up or down as needed, and to complete the transfer of its main forecasting systems by December 2027.
NOAA's use of cloud infrastructure grows to include weather prediction models | National Oceanic and Atmospheric Administration
NOAA ditches weather-predicting supercomputers for Google Cloud
https://www.theregister.com/hpc/2026/07/29/noaa-ditches-weather-predicting-supercomputers-for-google-cloud/5280697
The National Weather Service (NWS), a subsidiary of NOAA, uses a platform called the Weather and Climate Operational Supercomputing System (WCOSS) to run weather, ocean, and climate prediction models on a daily basis. Google Cloud will provide the primary high-performance computing infrastructure to power WCOSS.
NOAA has previously used HPE Cray supercomputers 'Dogwood' and 'Cactus,' located in data centers in Virginia and Arizona. Both systems were upgraded in 2023 and each boasted a computing performance of 14.5 petaFLOPS, but dedicated facilities presented challenges such as the time required to upgrade to new processors and respond to changes in computing demand.
At the new location, Google Cloud's virtual machines, 'H4D,' equipped with 5th generation AMD EPYC processors, will serve as the central computing infrastructure. H4D is designed to connect numerous virtual machines via a low-latency network, enabling large-scale numerical weather forecasting to be executed in parallel, much like a supercomputer.
The advantage of cloud computing is that it allows for increased processing capacity during periods of high computational demand, such as hurricane season, and can be scaled back when demand subsides. NOAA Administrator Neil Jacobs explains that it eliminates the traditional bottlenecks associated with on-premise systems and shortens the time it takes to translate new technologies from the research stage into actual forecasting operations.
The agreement with Google involves not only traditional numerical weather forecasting but also the introduction of AI-powered forecasting models. NOAA is developing the 'AI Global Forecast System (AIGFS)' using technologies such as Google DeepMind, which it claims will reduce the amount of computer computation required by 99.7% compared to conventional forecasting models, and shorten the forecast generation time from several hours to just a few minutes.
NWS is not only moving its forecasting calculation platform to the cloud, but also to the systems used by local weather stations to analyze data and issue warnings. In March 2026, Accenture and Booz Allen Hamilton were selected to develop and operate 'HIVE' and 'CIRRUS,' which run on the cloud.
NOAA explains that this transition will be one of the world's first examples of operating a national-scale, actionable numerical weather forecast on a public cloud. However, they will not immediately shut down their dedicated supercomputers; instead, they will gradually migrate forecasting models and related software from existing systems, aiming to complete the transition by the end of 2027.
Related Posts:
in Note, Posted by log1i_yk