China is racing Western AI labs to build competitive weather forecasting models, with three home-grown systems leading the effort: Fengwu (Shanghai AI Laboratory), Pangu (Huawei), and Fuxi (Fudan University). Fengwu reportedly outperformed Google DeepMind's GraphCast on roughly 80% of tested weather variables and predicted Typhoon Dolphin's landfall to within 30 minutes and 30km five days in advance. AI forecasters offer major speed and cost advantages over traditional numerical weather prediction, though they still lag on predicting storm intensity and face challenges with climate conditions outside their training data. The broader competitive landscape includes DeepMind's GenCast, Nvidia-backed FourCastNet, ECMWF's AIFS, and well-funded startups, making weather forecasting an unusually objective benchmark for AI rivalry since forecast accuracy is verifiable against real-world outcomes.
Questions this post answers
How accurate was China's Fengwu AI model at predicting Typhoon Dolphin's landfall?
Fengwu predicted Typhoon Dolphin's time and place of landfall to within 30 minutes and 30km (about 19 miles), five days before the storm hit the Chinese mainland. The prediction came from Sun Zhi of the firm Techwind and illustrates the model's potential for turning evacuation orders from guesswork into actionable logistics. Teams building or evaluating AI forecasting systems track competitive benchmarks like these on daily.dev.
How does China's Fengwu AI weather model compare to Google DeepMind's GraphCast?
Fengwu reportedly outperformed GraphCast across roughly 80% of the weather variables it was tested on and pushed skilful forecasts beyond the ten-day mark traditionally considered the edge of usefulness. These figures are pending independent scrutiny, but if confirmed, they suggest China has closed — and on some measures surpassed — the gap with Western AI weather models. Developers following the AI weather forecasting race find the latest model comparisons on daily.dev.