Google has rolled out its latest AI-powered weather forecasting model, WeatherNext 2, marking a major shift from experimental development to real-world deployment across its products. The new model has demonstrated strong accuracy and efficiency, generating forecasts up to 15 days in advance with hourly precision—capabilities that are now being integrated into Google Maps, Search, Gemini, and Pixel Weather. Peter Battaglia, senior director of research and sustainability at Google DeepMind, emphasized the transition from lab-based testing to widespread user access. “We’re taking it out of the lab and really putting it into the hands of users in more ways than we have before and sort of shedding off the experimental kind of designation because we have confidence that our forecasts are really quite effective and quite useful,” he said during a briefing with reporters. WeatherNext 2 runs eight times faster than its predecessor and achieves high accuracy across 99.9% of weather variables, including temperature, wind, and precipitation. It can produce hundreds of potential forecast outcomes from a single starting point in under a minute using a single Google TPU chip—tasks that traditionally take several hours on supercomputers using physics-based models. Unlike conventional forecasting systems, which simulate atmospheric physics in great detail and require massive computational resources, AI models like WeatherNext 2 learn patterns from vast historical weather data. This allows them to make predictions more efficiently while maintaining high precision. The breakthrough lies in Google’s use of a new approach called Functional Generative Network (FGN). Unlike earlier AI models that required multiple processing steps to generate a single forecast, FGN introduces controlled randomness—noise—into the model with each input. This enables the system to generate a wide range of possible future weather scenarios in a single pass, dramatically improving speed and flexibility. The model’s ability to deliver hyper-local, high-frequency forecasts is especially valuable for industries like energy, agriculture, transportation, and logistics, where precise timing can impact operations and decision-making. “We found that energy, agriculture, transportation, logistics, and customers in many other industries are quite interested in these one-hour steps. It helps them make more precise decisions relating to things that affect their business,” said Akib Uddin, a product manager at Google Research. In addition to public-facing features, Google is offering an early access program for enterprise customers who want to build custom weather models. Forecast data is also available through Google Earth Engine for geospatial analysis and BigQuery for large-scale data processing. Google is not alone in this space. Competitors including the European Center for Medium-Range Weather Forecasts, Nvidia, and Huawei are also advancing AI-driven weather prediction systems. However, Google’s integration of WeatherNext 2 across its ecosystem and its focus on scalability and accessibility position it as a key player in the next generation of weather forecasting.