Nvidia unveiled its Earth-2 family of AI models and tools for weather forecasting at the American Meteorological Society Annual Meeting. The stack provides the world’s first fully open, accelerated AI weather software, enabling scientists, developers, startups, enterprises, and government agencies to generate faster, cost-efficient forecasts.
The technology covers all stages of forecasting, from initial observation processing to 15-day global outlooks and local storm predictions. Models in the stack include Earth-2 Medium Range (Atlas architecture), Earth-2 Nowcasting (StormScope), and Earth-2 Global Data Assimilation (HealDA), complementing existing tools like Earth-2 CorrDiff and Earth-2 FourCastNet3.
Several organizations are already deploying Earth-2. Brightband uses Medium Range for daily global forecasts, while the Israel Meteorological Service reports a 90% reduction in compute time with CorrDiff. Energy companies such as TotalEnergies, Eni, and Southwest Power Pool, along with financial firms like S&P Global Energy and AXA, are leveraging the models for operational planning and risk assessment.
Earth-2 Medium Range and Nowcasting are available now via NVIDIA Earth2Studio, Hugging Face, and GitHub, with Global Data Assimilation expected later this year.
