China’s AI weather forecasting takes a major step forward

China’s push into AI weather forecasting just took a big step forward. Models built in China, including Shanghai AI Laboratory’s Fengwu, Huawei’s Pangu-Weather, and Fudan University’s Fuxi, can produce forecasts in minutes and at a fraction of the computing cost of traditional physics-based models that run on supercomputers, at least according to the teams behind them. That matters more as extreme weather keeps showing up more often.

Windy Download

The biggest shift is speed. Chinese researchers say that if forecasts come back faster, you can run more scenarios, update warnings more frequently, and maybe get alerts out earlier for typhoons, floods, and heatwaves. Those same researchers point to one Typhoon Dolphin forecast that put landfall within 30 minutes and 30 kilometers, five days in advance.

China is also moving hard against Western rivals such as Google’s GraphCast and GenCast, Nvidia-backed FourCastNet, the European Centre for Medium-Range Weather Forecasts, and new global models from the US National Oceanic and Atmospheric Administration. Chinese teams say Fengwu outperformed GraphCast on about 80% of the variables they tested, and they think it could eventually extend useful forecasting past the 10-day mark.

If weather alerts affect how you travel, farm, price insurance, or just plan your day, this is something to keep an eye on. Even so, experts say the older models still tend to do a better job with storm intensity and the rare, record-breaking events, and a lot of the splashier claims still need independent, peer-reviewed testing at scale.

You can’t download most of these systems yet. Still, the Chinese teams behind them are pushing open-source tools and outside partnerships, including projects with Thailand and Djibouti.