DeepMind Says Its AI Can Predict Hurricanes Earlier Than Everyone Else

DeepMind says its WeatherNext model can predict hurricanes earlier than other systems. The model forecasts both a storm’s track and its intensity. DeepMind plans to open-source WeatherNext. Coverage of the claim comes via Wired.

WeatherNext produces those track and intensity forecasts from lower-resolution weather data. That is unusual because finer grids are the usual path to sharper storm guidance. Researchers still do not fully understand how the model extracts useful signal from that coarser input.

For engineers and builders, the useful part is the product shape: one model that jointly handles path and strength, and that can run on data that is cheaper and more available than high-resolution fields. Open-sourcing the model also means teams can inspect weights, retrain on regional archives, and plug the outputs into warning pipelines without waiting on a closed vendor stack.

That puts WeatherNext in a market where operational centers and commercial weather providers still lean on physics-based numerical weather prediction, often at higher resolution, for official guidance. An earlier, dual-output forecast from lower-resolution inputs is a different tradeoff: less reliance on expensive fine-grid runs, more reliance on a learned system whose internal reasoning is not yet clear.

What to watch next is the open-source release itself: whether the published artifacts match the track-and-intensity claims on independent storms, how well the model holds up outside the cases DeepMind highlighted, and whether operational groups will treat it as a primary forecast tool or as a fast ensemble member beside existing NWP. Until the how is better understood, the practical stance is to validate hard on holdout events before wiring it into automated alerts.