Artificial intelligence (AI) is often criticized for the energy consumption of data centers, which leads to increased use of fossil fuels and climate change. However, the same AI helps meteorologists predict hurricanes much more accurately - some of the hardest phenomena to forecast, which are becoming increasingly dangerous in a warming climate.
The US National Hurricane Center (NHC) began using the Google DeepMind model last year and is working with the company on its successor. The work came to the fore this year: an intense El Nino (an oscillation of ocean temperature that affects weather) led to a record number of hurricanes in the Pacific Ocean, and Hurricane "Isaias" struck the southeastern United States last month as a powerful storm.
Ferran Alet, a DeepMind research scientist who leads the cyclone project, explained that Google does not have a separate model for hurricanes - the same model used for everyday weather is used, but at the limit of its capabilities. "We view extreme weather as a challenge worth striving for, because it is the hardest thing for us, and the rest will follow," he said.
Before AI, the NHC used several types of models. "Dynamic" models solve equations of fluid motion and atmospheric thermodynamics on supercomputers. Statistical models rely on historical relationships between storm behavior and variables such as location and date. Hybrid models combine physics and statistics. AI models, like statistical ones, learn from large data, but on a much larger scale and are able to capture more complex patterns.
A DeepMind paper published in the journal Nature in August showed that the model on average gives forecasters an extra day of accurate forecasting - a three-day forecast was as good as previous models were for the next two days. For "Isaias", the model recognized that the storm could become a tropical cyclone a full week before it formed. It also generates 1,000 possible scenarios for each cyclone and forecasts precipitation along with track and intensity (storm strength).
"The NHC began using AI models in operational work at the end of June 2025, and the Google DeepMind model was added to some auxiliary tools for track and intensity forecasting in 2026," said Dan Brown, chief of the NHC's operations branch. The center also uses models from the European Centre for Medium-Range Weather Forecasts and the National Weather Service and has an agreement with Google to develop its models. "During 'Isaias', the AI models performed well on track," Brown noted. "They seem to quickly pick up changes in complex steering flows."
Brown singled out DeepMind and the NHC's own model called HCCA as some of the best tools for track and intensity. But the technology is not without limitations: individual models can "change track or intensity forecasts sharply from cycle to cycle," and specialists have to rely on experience to decide which models better describe a particular storm and to combine them. He also recalled that physical models created the foundation on which AI models were trained. "As for the hurricane specialist? They are becoming more important than ever. An algorithm can show the probability of a track, but it does not look at aircraft and satellite data to determine the current structure of the storm," Brown said.
The increase in forecast accuracy comes amid rapid data center construction in the United States that is changing the structure of the power grid: new centers, according to a Cleanview analysis, will mostly run on natural gas, increasing the carbon footprint and worsening climate change. And that, in turn, makes hurricanes more powerful, brings more precipitation and slows their movement, amplifying the consequences.