WeatherNext

WeatherNext is a weather-prediction AI built by Google DeepMind. Traditional forecasting works by running physics equations from scratch on supercomputers the size of a three-story building. WeatherNext, by contrast, is trained on roughly 40 to 50 years of past weather records, and works more like a veteran weather forecaster who recognizes that "this kind of pattern usually turns into that." Because it doesn't need to solve equations from scratch, it requires far less computation, meaning a single chip can run forecasts instead of a massive supercomputer.

The model came out of a failure. The first version, released in 2023, was good at predicting a storm's path but poor at predicting how strong it would become. That same year, Hurricane Otis suddenly intensified into a top-category storm overnight and slammed into Mexico, catching every physics-based model off guard. That incident pushed Google DeepMind to build a version specialized in catching rapid intensity changes in cyclones. In 2025, that version successfully warned in advance that Hurricane Melissa would rocket from the weakest category to the strongest within a single day.

News coverage, under headlines like "Google DeepMind's AI catches hurricane rapid intensification a day in advance," explains that WeatherNext predicted Hurricane Melissa's jump from Category 1 to Category 5 a full day ahead. A common misunderstanding is that this model fully replaces traditional supercomputer-based weather simulation. In reality, the two approaches are used together, with WeatherNext playing a complementary role by catching signs of sudden change that physics-based simulations tend to miss.