Nvidia has introduced a new suite of open models and tools for AI-powered weather forecasting, unveiled on Monday at the American Meteorological Society Annual Meeting.
The Earth-2 family represents the world’s first fully open and accelerated weather AI software stack, designed to make advanced weather and climate forecasting more accessible to scientists, startups, developers, enterprises, and government agencies worldwide.
This new technology speeds up every stage of the forecasting process, from handling raw observational data to producing 15-day global outlooks or localized storm forecasts. By relying on AI rather than traditional physics-based simulations run on supercomputers, Earth-2 significantly reduces computational time and operating costs.
The newly released models include Earth-2 Medium Range, powered by the Atlas architecture; Earth-2 Nowcasting, built on StormScope; and Earth-2 Global Data Assimilation, driven by HealDA. These additions expand Nvidia’s existing Earth-2 stack, which already features models such as Earth-2 CorrDiff and Earth-2 FourCastNet3.
A growing number of organizations are already adopting the technology. AI weather platform Brightband is using Earth-2 Medium Range to deliver daily global forecasts, while the Israel Meteorological Service is operating Earth-2 CorrDiff and reports a 90% reduction in computing time compared with conventional forecasting methods.
Energy companies including TotalEnergies, Eni, GCL, and Southwest Power Pool are assessing or deploying Earth-2 models to enhance operational forecasting. In the financial sector, firms such as S&P Global Energy and AXA are applying the technology for risk analysis and scenario modeling.
Earth-2 Medium Range and Earth-2 Nowcasting are now available through NVIDIA Earth2Studio, Hugging Face, and GitHub. Earth-2 Global Data Assimilation is expected to be released later this year.
