Google has introduced its new AI weather forecasting model, WeatherNext 3, tailored specifically for the energy market.

This latest version incorporates new variables that significantly impact power generation assessments and risks for electrical grids, such as wind speed at 100 meters, cloud cover, and the amount of sunlight reaching the surface.

Compared to its predecessor, WeatherNext 2, which operated on a 25 km grid and updated every six hours, WeatherNext 3 is more detailed and responsive. It provides a global forecast updated hourly and offers a 5 km resolution for ground variables like temperature and humidity, and an 11 km resolution for precipitation.

According to Google, rather than relying solely on simulations, WeatherNext 3 learns directly from meteorological station data and recent satellite images. This approach has reduced information latency from seven hours to between three and four hours, as noted by Ilan Price, a senior researcher at Google DeepMind.

Accuracy of air temperature forecasts by AI models at 850 hPa. Source: Brightband.

The company reported enhancements in precipitation estimation accuracy:

  • up to 60% compared to NASA IMERG;
  • up to 30% relative to MRMS;
  • up to 10% compared to rain gauges in early forecast horizons.
Source: Google.

However, training solely on historical patterns could pose challenges during unprecedented weather anomalies, while traditional physical models are based on fundamental atmospheric laws.

The model has been integrated into Google Search, Gemini, Google Maps, and the Google Maps Platform Weather API. Corporate clients can access it via BigQuery, Earth Engine, and Google Cloud Storage, eliminating the need for clients to deploy the software themselves.

At launch, Google did not disclose the cost of a corporate subscription for WeatherNext 3.

Market for Weather Data

According to the S&P Global Market Intelligence US Grid Outlook 2026, solar generation and storage are expected to be the primary sources of new capacity in the U.S. for 2026, with 51.2 GW and 25.7 GW, respectively, from over 90 GW planned.

The demand for timely information on weather changes is driven by the high cost of errors in procurement planning.

For instance, if wind generation is underestimated, additional electricity must be purchased at the last minute—often from gas plants. Conversely, if forecasts are too high, wind and solar outputs may be curtailed because the grid cannot accommodate the entire volume.

Key competitors to Google in the weather data market for the energy sector include Vaisala, Solcast, DNV's WindGEMINI, IBM HyperWatch, and the Swiss company Jua.

For Google, the launch of WeatherNext 3 is not just about offering a new product; it also serves as a tool for optimizing its own costs. The corporation purchases renewable energy to power its computing clusters, and accurate generation forecasts directly influence its financial performance.

In March 2026, researchers from the UK and Canada developed an AI model called Aardvark Weather for weather forecasting.