China’s weather agency unveils 2035 AI‑driven Earth system forecast plan

This strategy document is nominally about weather and climate, but it’s also a quiet milestone in how a major government plans to industrialize AI for high‑stakes scientific forecasting. China’s meteorological agency is explicitly committing to a hybrid stack: physics‑based Earth system models plus AI models trained on massive observational datasets, with the aim of building a kilometer‑scale global forecast system by 2035. That kind of program creates a long, predictable funding stream for large models tightly coupled to HPC infrastructure, which in practice means more talent, compute, and institutional experience flowing into AI‑for‑science. ([ts.cn](https://www.ts.cn/xwzx/gnxw/202512/t20251218_32118587.shtml))

For the AGI race, these investments matter because they push AI beyond chatbots into domains where long‑horizon reasoning, data efficiency and robustness under distribution shift are non‑negotiable. Training models that can learn from and interact with coupled atmosphere–ocean–land systems is a very different challenge from next‑token prediction on web text; success here will feed back into better architectures, training curricula and evaluation methods. Strategically, it also signals that China sees AI weather and climate modeling as a dual‑use capability: critical for disaster resilience and agriculture, but also a proving ground for large‑scale, safety‑critical AI deployment. That combination of political backing, national‑level coordination and explicit AI–HPC fusion is exactly the kind of institutional muscle that can compress timelines if the technical pieces fall into place.