MANY MINDEDNOAA Plans to Move National Weather Forecasting to Google Cloud
NOAA has selected Google Cloud to run its weather forecasting infrastructure, saying it will be the first national weather prediction center to operate on the commercial cloud. It plans to move its Weather and Climate Operational Supercomputing System, run by the National Weather Service, to the cloud by December 2027. The prior contract, held by General Dynamics, managed two HPE Cray supercomputers named Dogwood and Cactus that together delivered nearly 14 petaflops.
The replacement uses Google Cloud H4D virtual machines built on AMD Epyc processors. Google says those instances can cost as low as 3 cents per core-hour without long-term commitments. NOAA also gains access to Google DeepMind AI tools to build an AI Global Forecast System, which Google says can produce accurate forecasts using 99.7% fewer compute cycles, in minutes rather than hours. Earlier, in March, NOAA hired Accenture and Booz Allen Hamilton to build cloud software named HIVE and CIRRUS.
The shift matters because it treats frontier high-performance computing as a rentable commodity rather than a fixed capital asset. If a national weather service can forecast on the same on-demand infrastructure anyone can rent, the barrier to serious public science drops. The UK Met Office is making a similar move to Microsoft Azure in a hybrid setup, suggesting a pattern.
The caveats are worth keeping. The performance and pricing claims come from Google, and the migration is a planned transition targeted for late 2027, not a completed one. NOAA Administrator Neil Jacobs framed the move as eliminating on-premise bottlenecks, but the results will only be clear once operations actually run there.
Source: The Register