NVIDIA has unveiled a new family of open models, libraries and frameworks under its Earth-2 initiative, making advanced AI-powered weather and climate prediction accessible to scientists, developers, enterprises and government agencies worldwide.

Announced at the American Meteorological Society’s Annual Meeting, the NVIDIA Earth-2 open software stack is the world’s first fully open, accelerated platform for weather and climate AI, covering every stage of forecasting from processing raw observation data to generating 15-day global forecasts and hyper-local storm predictions.

The open technologies include pretrained models, customisation recipes, inference libraries and developer frameworks that significantly reduce the computational cost and time required for high-resolution forecasting. Traditionally dependent on large supercomputers running physics-based simulations, weather prediction can now be performed more efficiently using AI-driven models, enabling wider adoption across countries and industries.

By allowing organisations to run, fine-tune and deploy production-ready weather AI on their own infrastructure, Earth-2 aims to accelerate forecasting accuracy, enhance collaboration and advance climate science at scale.

Developers across energy, finance and public sector domains are already adopting the platform. Users include national weather agencies such as the U.S. National Weather Service, Taiwan’s Central Weather Administration and Israel Meteorological Service, as well as companies including TotalEnergies, Eni, The Weather Company, AXA and S&P Global Energy.

NVIDIA said Earth-2 represents a major step toward faster, more precise and accessible climate intelligence worldwide.