AI weather forecasting enters the energy market as Google targets grid operators with WeatherNext 3

Details

Google DeepMind and Google Research released WeatherNext 3 on September 3, entering the commercial energy forecasting market for the first time. The model generates hourly global forecasts at up to 5km resolution — up from WeatherNext 2's 25km grid with 6-hour refresh cycles. Energy-specific outputs include wind speed at 100m (standard turbine hub height), cloud cover, and surface solar irradiance. Google names Vaisala, Solcast, and DNV as the incumbents it now competes with. Enterprise distribution is designed for zero-friction adoption: the forecast data is queryable in BigQuery and Google Earth Engine, or downloadable in bulk from Google Cloud Storage, with no model deployment required. This removes an integration barrier that competing vendors face when selling into large utility buyers. The timing aligns with structural grid changes. S&P Global's US Grid Outlook 2026 projects over 90GW of new US capacity this year, led by solar (51.2GW) and storage (25.7GW). Each added gigawatt of weather-dependent generation raises the cost of forecast errors: underestimates force operators to buy expensive standby gas power; overestimates result in paid wind and solar curtailments. Deloitte projects data center electricity demand reaching 176GW by 2035 — five times the 2024 level — adding load-side complexity that makes accurate short-term forecasting more commercially valuable. WeatherNext 3 also powers Google Search, Gemini, and Google Maps weather, giving the model broad consumer-facing distribution while Google establishes itself in an energy data market it had not previously entered commercially.