NOAA Google Cloud deal shifts U.S. weather forecasting off agency supercomputers
NOAA plans to move National Weather Service forecasting to Google Cloud by December 2027, replacing government-run supercomputer operations.
By Renata Fuchs · Policy Reporter
· 3 min read
NOAA has selected Google Cloud to run the infrastructure behind U.S. weather forecasting, shifting National Weather Service operations away from agency-hosted supercomputers. The NOAA Google Cloud move is scheduled to put the Weather and Climate Operational Supercomputing System, along with software used to generate NWS weather data for analysis, onto commercial cloud infrastructure by December 2027.
The value of the contract was not disclosed. NOAA said it will be the first national weather prediction center to operate on the commercial cloud, although the U.K. Met Office is also moving its forecasting system to Microsoft Azure in a hybrid arrangement.
The change is a notable defection from the traditional high-performance computing model used by weather agencies. NOAA’s prior weather-computing operations were managed by General Dynamics and ran most recently on HPE Cray systems in data centers in Virginia and Arizona. Those systems, named Dogwood and Cactus, delivered nearly 14 petaflops for forecasting workloads.
Why is NOAA moving weather forecasting to Google Cloud?
NOAA is betting that cloud-based high-performance computing will let it add capacity during peak demand and reduce it when demand falls, rather than sizing around fixed in-house systems. NOAA Administrator Neil Jacobs said in a statement that cloud HPC should speed the path from research into operations by removing bottlenecks associated with on-premises infrastructure.
The seasonal logic is straightforward. Forecasting demand rises during periods such as tropical storm season, when agencies need more compute cycles for models and warning systems. A cloud setup gives NOAA a way to rent more capacity during those windows, then scale back during quieter periods, assuming the workloads and data flows perform as expected.
The move also gives NOAA access to Google’s DeepMind tools under the contract. NOAA plans to use them as part of its AI Global Forecast System, described as its first AI-driven weather forecasting system. The agency says the system is intended to produce accurate forecasts with 99.7% fewer compute cycles and generate forecasts in minutes rather than hours. NOAA did not provide a production benchmark in the details reported, so the claim remains a target rather than an independently measured operating result.
The cloud migration is not limited to the central forecasting models. The National Weather Service has also been updating software downstream from its Global Forecast System and Global Ensemble Forecast System so those tools can run in the cloud. In March, NWS awarded Accenture and Booz Allen Hamilton contracts to develop cloud-based software, known as HIVE and CIRRUS, for field offices to analyze data and distribute alerts. Those systems are intended to replace current in-house software used for those functions.
What infrastructure will Google use for NOAA weather models?
Google plans to run the work on Google Cloud H4D virtual machines built with AMD Epyc processors. Google classifies the instances as virtual machines because they operate under a hypervisor that ties into its networking and orchestration systems. The company says that setup allows the instances to be synchronized for large jobs in a way similar to supercomputing clusters.
For organizations running HPC jobs, Google offers Cluster Toolkit for cluster deployment, Cluster Director for cluster maintenance and Google Cloud Batch for queuing, scheduling and resource provisioning. Google says H4D instances can be accessed for as little as 3 cents per core-hour without long-term commitments.
For the broader HPC market, NOAA’s decision is a signal that at least some government-scale scientific workloads are moving from owned systems to cloud capacity. Weather prediction has been one of the workloads that justified government-funded supercomputing deployments. If NOAA’s migration works at operational scale, cloud providers gain a reference case in a field where uptime, repeatability and data throughput are harder to market around than generic AI capacity.
This story draws on original reporting from The Register.
