NWS shifts to cloud computing to support AI weather models

For decades, the backbone of American weather forecasting has been two massive, physical supercomputers housed in dedicated data centers. These on-premises systems, known as the Weather and Climate Operational Supercomputing System, have reliably powered the numerical models that forecasters depend on. However, National Weather Service officials are now prioritizing flexibility over raw, static power. As reported by GN technics/ai (en-US), the agency is using the expiration of its current hardware as a catalyst to modernize its infrastructure, moving high-performance computing into a cloud environment. This change is not merely an IT upgrade; it is a structural adjustment to handle a landscape where computing requirements are shifting rapidly due to new technologies.

The primary driver behind this transition is the rise of artificial intelligence in weather prediction. Traditional numerical weather prediction requires a specific, massive amount of processing power, often measured in teraflops, but it follows a predictable computational path. AI-driven models, however, have different demands. They may require less raw compute for some tasks but need the ability to scale and adjust instantly. David Michaud, director of NWS central operations, explained that the goal is to balance computing power with adaptability. In a fixed on-premises setup, adding capacity for a sudden surge in AI processing is slow and expensive. In the cloud, resources can be allocated on demand, allowing the agency to pivot its computing focus as new AI methods prove effective.

Flexibility over fixed hardware

The trade-off for this flexibility is a move away from self-contained infrastructure. By shifting to the cloud, the NWS is becoming dependent on external providers for its most critical operational tools. The agency has selected Google Cloud as the primary provider for this high-performance infrastructure. This decision reflects a broader trend in government agencies to adopt cloud solutions for their agility. The catch, however, is that this creates a new layer of complexity in managing data security and operational continuity. While the cloud offers the ability to quickly shift the balance of compute resources to accommodate new AI models, it requires a different skill set in management and a higher level of trust in third-party systems.

This shift also allows for a more streamlined workflow for meteorologists and researchers. Currently, the agency is simultaneously moving weather data to cloud-based platforms like NWS HIVE and NWS CIRRUS. These platforms enable forecasters to access crucial data via mobile devices, away from their home offices. Combining this data access with cloud-based supercomputing creates a more integrated technology footprint. The ability to run models and access data in the same environment reduces bottlenecks and allows for faster iteration. This is particularly important during severe weather events, where the agency needs to manage normal workloads while simultaneously processing high-demand AI predictions.

AI models enter operational use

The NWS is not waiting for AI technology to mature in isolation; it is already integrating these tools into its operational models. The agency has deployed several AI-fueled systems, including AIGFS and AIGEFS, which use less compute to produce forecasts and provide probabilistic information to forecasters, respectively. There is also a hybrid system called Hybrid-GEFS that combines traditional numerical prediction with AI enhancements to improve accuracy. Richard Bandy, director of the NWS’s Meteorological Development Laboratory, noted that moving to the cloud actually returns this technology to its environment of origin. Many of these AI models were initially developed in cloud environments by partners like Google’s DeepMind. By moving the operational backbone to the cloud, the NWS is aligning its production infrastructure with the development environment, reducing friction in the deployment of new forecasting capabilities.

Despite the promise of AI, there are significant limitations. Artificial intelligence in weather prediction is still in its early stages, particularly regarding its performance in extreme weather situations. While it shows promise for improving general forecasts, it is not yet a complete replacement for traditional methods. The cloud transition is designed to accommodate this uncertainty. If AI capabilities develop quickly, a fixed on-premises system with a finite set of compute might not be able to shift its balance to accommodate the new demands. The cloud, however, allows for rapid adjustment. This adaptability is the core stake for the reader: a weather service that can evolve its tools as fast as the technology changes, rather than being locked into a decade-old hardware architecture.