Google’s New Hurricane Model Was Breathtakingly Good This Season

Why It Matters

The superior accuracy and faster turnaround of Google’s AI model could reshape hurricane prediction, reducing reliance on costly supercomputer‑based physics models and improving public safety and insurance risk assessments. Its success signals a broader shift toward data‑driven forecasting across the meteorological industry.

Summary

Google DeepMind’s Weather Lab AI model, released in June, outperformed the US National Weather Service’s Global Forecast System (GFS) and even the official National Hurricane Center forecasts in the 2025 Atlantic hurricane season. Across 13 named storms, the DeepMind model’s five‑day track error averaged 165 nautical miles versus 360 nautical miles for the GFS, and it also delivered superior intensity forecasts. The AI model consistently beat consensus products such as TVCN and HCCA, while the GFS lagged dramatically, raising questions about its recent core upgrade. Analysts expect AI‑driven weather models to become central to future forecasting.