Google updates its weather forecasts with a new AI model (www.theverge.com)

Google announced WeatherNext 2, a new AI weather model that will power forecasts across Search, Maps, Gemini and Pixel Weather and be offered to enterprise customers via early access, BigQuery and Earth Engine. Google says the model is eight times faster than its prior system and more accurate — claiming improved performance on 99.9% of variables such as temperature and wind. WeatherNext 2 can produce hundreds of possible outcomes in under a minute using a single TPU, a task that would traditionally take hours on supercomputers running physics-driven simulations. The technical leap comes from a Functional Generative Network (FGN) approach that injects targeted noise into inputs so many forecast realizations are generated in one step instead of repeating heavy computation. That efficiency lets WeatherNext 2 deliver hourly forecasts and projections up to 15 days out, which Google frames as valuable for energy, agriculture, transportation, logistics and other industries needing fine-grained, time-resolved predictions. The move signals Google is transitioning AI-based forecasting from experimental research into product features, and it puts the company in direct competition with others (ECMWF, Nvidia, Huawei, etc.) racing to commercialize generative AI for weather.