NVIDIA Earth-2: Open Weather AI Revolution

1/26/2026

NVIDIA has unveiled a transformative set of tools for the meteorological community at the American Meteorological Society’s Annual Meeting: the NVIDIA Earth-2 family of open models. This release marks the world’s first fully open, accelerated software stack designed specifically for weather and climate AI. Historically, high-fidelity weather forecasting relied heavily on massive supercomputers running complex physics-based models—resources typically limited to large government agencies. NVIDIA Earth-2 changes this paradigm, making advanced weather and climate prediction more accessible than ever to scientists, startups, developers, enterprises, and government agencies worldwide. From processing initial observation data to generating 15-day global forecasts or localized storm alerts, Earth-2 technologies accelerate every stage of the forecasting pipeline. https://blogs.nvidia.com/wp-content/uploads/2026/01/storm_scope_007_high_30MB.mp4 Breakthrough Architectures: Atlas, StormScope, and HealDA The new open models are powered by three distinct architectures. First, Earth-2 Medium Range utilizes the Atlas architecture to enable high-accuracy predictions for up to 15 days in advance. Covering over 70 weather variables—such as temperature, pressure, wind, and humidity—it outperforms leading open models on standard industry benchmarks. https://blogs.nvidia.com/wp-content/uploads/2026/01/forecast_dramatic_multivar_new_30MB.mp4 Second, Earth-2 Nowcasting is powered by the StormScope architecture. Using generative AI trained on satellite and radar data, it predicts the evolution of realistic cloud and rainfall systems. This model generates kilometer-resolution, zero-to-six-hour forecasts of local storms and hazardous weather in just minutes. By simulating storm dynamics directly, it becomes the first AI model to outperform traditional physics-based methods for short-term precipitation forecasting. https://blogs.nvidia.com/wp-content/uploads/2026/01/combined_rotate_4k_30s_30MB-1.mp4 The third component, Earth-2 Global Data Assimilation, is powered by the HealDA architecture. It produces the vital "initial conditions" for weather prediction—snapshots of the current atmosphere at thousands of locations globally. While creating these snapshots typically takes hours on supercomputers, HealDA accomplishes it in seconds on GPUs. When paired with Earth-2 Medium Range, this creates a highly skillful, fully AI-driven forecasting pipeline. Adoption Across Industries Developers across various industries are already harnessing Earth-2. The Israel Meteorological Service uses Earth-2 CorrDiff to generate high-resolution forecasts up to eight times daily. Director Amir Givati noted a 90% reduction in compute time at 2.5-kilometer resolution compared to classic numerical models, with the AI model proving superior during recent storm verifications. https://blogs.nvidia.com/wp-content/uploads/2026/01/demo_fcst-1.gif In the energy sector, TotalEnergies is evaluating Earth-2 Nowcasting to improve decision-making; Emmanuel Le Borgne describes these models as groundbreaking for managing short-term risks in energy systems where local impacts matter. GCL, a major solar material producer, reports that Earth-2 provides more accurate prediction data at a lower cost than traditional methods, enhancing their photovoltaic power generation forecasts. Additionally, Southwest Power Pool, in collaboration with Hitachi, uses the models to improve wind forecasting and grid reliability. In finance, AXA leverages the stack to generate thousands of hypothetical hurricane scenarios for advanced risk assessment.