Google Unveils WeatherNext 2: 8x Faster AI Weather Forecasting

Lightning-Fast Predictions at Scale

Google DeepMind and Google Research released WeatherNext 2 on November 17, 2025, delivering weather forecasts eight times faster than its predecessor while improving accuracy across 99.9% of meteorological variables. The AI model generates hundreds of possible weather scenarios from a single starting point in less than 60 seconds using just one tensor processing unit (TPU), a task that would take hours on traditional supercomputers. The technology is already powering weather information across Google Search, Maps, Gemini AI assistant, and Pixel Weather apps.

Novel AI Architecture Enables Breakthrough

At the core of WeatherNext 2 sits a Functional Generative Network (FGN), a new AI approach that injects strategic "noise" into the model architecture to create physically realistic forecasts. The model trains only on individual weather elements like temperature or humidity at specific locations, yet learns to predict complex interconnected systems such as regional heat waves or wind farm power output. This capability addresses a critical challenge in meteorology: capturing worst-case scenarios and the full range of weather possibilities essential for planning decisions.

Enterprise and Research Access

Google is democratizing access through multiple channels, making WeatherNext 2 forecast data available in Earth Engine and BigQuery. An early access program on Google Cloud's Vertex AI platform enables businesses to create customized forecasting models. Weather agencies have already utilized the technology for experimental cyclone predictions, making critical decisions based on multiple scenario forecasts.

Higher Resolution, Greater Detail

The model delivers forecasts at resolutions down to one-hour intervals, generating four six-hour forecasts daily. This temporal granularity represents a significant improvement over previous systems, providing meteorologists and decision-makers with more detailed information for time-sensitive applications. WeatherNext 2 demonstrates particular strength in forecasting crucial variables including wind speed and direction, precipitation, and atmospheric pressure across lead times ranging from zero to 15 days.

From Lab to Real-World Impact

Google's integration strategy spans consumer applications, enterprise platforms, and research tools, positioning WeatherNext 2 for widespread adoption across sectors. Global supply chains, aviation operators, and individual users all gain access to more accurate predictions for route planning, safety decisions, and daily activities. The company has committed to continued advancement, including integration of new data sources and expanding access further to accelerate scientific discovery and empower researchers, developers, and businesses worldwide.