The Earth-2 suite consolidates multiple AI-powered weather models, including Earth-12 Medium Range for 15-day forecasts over 70+ variables, Nowcasting with kilometer-scale local six-hour predictions, Global Data Assimilation for atmospheric initial conditions, CorrDiff to downscale global data up to 500x faster, and FourCastNet3 delivering wind, temperature, and humidity forecasts at up to 60x the speed of traditional methods.

In addition to CorrDiff and FourCastNet3, the Earth-2 lineup features the PhysicsNeMo framework for training and fine-tuning AI-physics models, enabling broader experimentation within the same platform.

By leveraging GPUs instead of CPU clusters, Earth-12 can cut processing times from hours to seconds, democratizing access to high-resolution forecasts for researchers, weather services, energy traders, and public agencies.

This development fits a broader trend of AI-enabled, locally deployable weather intelligence aimed at improving readiness and decision-making in a warming world.

The system is designed to accelerate forecasting across all stages, from processing observational data to producing both global and localized storm predictions.

Energy companies are among the users, leveraging Earth-2 to improve grid reliability and photovoltaic predictions.

CorrDiff uses generative AI to downscale coarse continental predictions to high-resolution regional forecasts, delivering results up to 500x faster than traditional methods.

Real-world adoption includes BrightBand using Earth-12 Medium Range for daily forecasts, the Israel Meteorological Service achieving substantial compute-time reductions with CorrDiff and improved precipitation verification, and energy majors like TotalEnergies, Eni, and GCL employing Nowcasting and FourCastNet to sharpen short-term risk awareness and forecast optimization.

NVIDIA announces Earth-12 as an open-access AI weather and climate software stack designed to accelerate forecasting and broaden access beyond large national institutions.

NVIDIA positions Earth-12 as a catalyst for global collaboration in climate tech by making models openly available via Earth2Studio, GitHub, and Hugging Face, aiming to accelerate scientific progress and planetary resilience.

Earth-12 can generate thousands of extreme-weather scenarios for climate risk assessment and insurance modeling, with users including AXA and S&P Global Energy.

FourCastNet3 delivers wind, temperature, and humidity forecasts that outperform conventional ensembles and operate up to 60x faster.