NOAA Moves Weather Supercomputing To Google Cloud By 2027
The Register reported that NOAA will move its National Weather Service supercomputing system to Google Cloud by December 2027, replacing HPE Cray machines with H4D virtual machines while leaving cutover and reliability testing details unpublished.
NOAA's weather-forecasting infrastructure is transitioning from government-managed supercomputers to Google Cloud by December 2027, converting a public high-performance computing workload into a commercial-cloud deployment.
The National Oceanic and Atmospheric Administration selected Google Cloud for the National Weather Service system that generates weather data for analysis, as reported on July 29.
This shift replaces a model built around dedicated HPE Cray machines with one based on cloud capacity that can scale up during tropical storm season and scale down during calmer periods.
The tension lies not only in cloud migration but also in whether a forecasting workload that previously depended on named government supercomputers can maintain operational reliability while moving to virtualized infrastructure.
NOAA Targets December 2027 For Weather Computing Move
The Weather and Climate Operational Supercomputing System will transition to Google Cloud along with the software used to generate NWS weather data.
General Dynamics managed the previous machines, while the most recent systems were HPE Cray supercomputers named Dogwood and Cactus in Virginia and Arizona.
The older systems were reported to have almost 14 PFlops of weather-prediction capacity.
NOAA Administrator Neil Jacobs stated that cloud-based high-performance computing would accelerate the transition of research into operations by removing bottlenecks tied to on-premise systems.
Google H4D VMs Replace Dedicated Cray Machines
Google plans to utilize H4D virtual machines built on AMD Epyc processors.
The H4D setup combines virtualization with Google's network and orchestration layer, allowing large jobs to run in a synchronized cluster model rather than on a single owned supercomputer.
The operational case relies on elasticity.
Jacobs linked the cloud model to the ability to ramp compute cycles during tropical storm season and reduce them when weather demand is lower.
This represents a different procurement and operations pattern compared to maintaining fixed supercomputer capacity in government-funded facilities.
AI Forecasting Enters The Same Contract
The contract also provides NOAA access to Google's DeepMind tools for an AI Global Forecast System.
This system is positioned alongside Google's claim that it can produce accurate forecasts using 99.7% fewer computer cycles and minutes of processing time instead of hours.
NWS software work is already moving downstream.
In March, the agency awarded contracts to Accenture and Booz Allen Hamilton for cloud-based HIVE and CIRRUS tools that field offices can use to analyze data and issue alerts, replacing in-house software previously used for those tasks.
Google prices entry-level H4D use at 3 cents per core-hour without requiring a long-term commitment.
The same tooling stack includes Cluster Toolkit for deployment, Cluster Director for cluster maintenance, and Google Cloud Batch for queuing, scheduling, and resource provisioning.
Operational Proof Moves To Forecast Reliability
The migration places a national weather workload within a commercial-cloud operating model that must support seasonal spikes, field-office alerting, and research-to-operations transfer.
It also provides Google with a public-sector reference for cloud-based HPC at a time when weather modeling, AI forecasting, and scientific computing are all competing for specialized processors and networked capacity.
NOAA has not disclosed the final cutover sequence, service-level terms, or independent reliability tests for the 2027 migration.
