Key developments: AI models are now rivaling or surpassing national supercomputer forecasts. Norfolk’s WeatherQuest reports AI sometimes beats Met Office supercomputers while using far less power. The ECMWF’s AI ensemble produces 51 scenarios over ten times faster than its physics‑based system, consumes about 1,000× less energy, and lifts accuracy up to 20%. Google DeepMind’s WeatherNext 2 delivers 15‑day outlooks eight times faster than its predecessor, giving cyclone tracks a day‑earlier warning. Recent operational launches—ECMWF’s AI ensemble (July 2025) and its AIFS system (Feb 2025)—show AI is moving from research to daily use.
Why it matters: Faster, cheaper forecasts enable more ensemble runs, giving decision‑makers richer probabilistic information while slashing compute costs. The reduced energy footprint addresses climate concerns tied to traditional supercomputing. Early‑warning gains, such as AI‑driven cyclone tracking arriving a day sooner, can save lives and property. Moreover, AI’s ability to predict extreme rainfall days ahead—demonstrated in a Science Advances study—offers crucial lead time for flood preparedness.
What to watch next: • Expansion of AI ensembles across more variables and higher resolutions, testing whether the 20% accuracy boost scales globally. • Integration of AI with real‑time observation streams (satellites, drones) to improve data assimilation. • Ongoing validation of AI performance on rare, high‑impact events, ensuring models remain reliable as climate extremes increase. • Adoption by national services, following the UK Met Office and US NHC’s growing reliance on AI guidance for hurricane forecasts. [[1, 2, 3]]