Nvidia has open sourced three GPU-powered AI weather models designed to replace supercomputer forecasting, aiming to democratise high-performance meteorology while cutting energy use and operational costs.
Nvidia has released three open source AI models for weather forecasting, positioning them as significantly faster and more energy-efficient alternatives to traditional supercomputer-based systems. The announcement was made at the American Meteorological Society Conference in Houston via Nvidia’s Earth-2 platform, marking a major push to open-source operational-grade weather forecasting infrastructure.
The models are designed to replace or augment physics-based numerical weather prediction using GPU-accelerated AI workflows. Earth-2 Medium Range delivers forecasts up to 15 days in advance, analysing more than 70 atmospheric variables including temperature, air pressure, wind, and humidity. Built on Nvidia’s Atlas architecture, the company claims it processes more variables than Google DeepMind’s GenCast model.
Earth-2 Nowcasting targets short-term forecasting up to six hours, using generative AI to predict satellite and radar data. Nvidia says it is the first AI model to outperform traditional physics-based systems in simulating storm dynamics. The third model, Earth-2 Global Data Assimilation, generates precise initial atmospheric conditions in seconds on GPUs, a process that typically takes hours on supercomputers. It is based on the HealDA architecture and will be released later in 2026.
Earth-2 Medium Range and Earth-2 Nowcasting are available via Nvidia Earth2Studio, Hugging Face, and GitHub. Performance claims include reductions in computation from hours to seconds, with the CorrDiff model cited as up to 500 times faster and 10,000 times more energy-efficient than CPU-based systems.
The open source models are already being tested by the US National Weather Service, Taiwan’s Central Weather Administration, and Israel’s Meteorological Service, while energy and insurance firms including TotalEnergies, AXA, and JBA Risk Management are using them for risk analysis. According to Alex Philp of MITRE Corporation, the approach marks a “turning point for weather forecasting.”
