Google's new weather model forecasts hourly, and it also forecasts wind farms

Google started rolling out WeatherNext 3 today. The model reads live satellite observations and produces a forecast every hour, against every six hours for the version it replaces, and it resolves temperature and moisture down to five kilometres where the old model worked on a 25-kilometre grid. Google puts the improvement in precipitation forecasts at up to 50% a day or more ahead.

What changed under the hood

Traditional forecasting runs the physics of the atmosphere on supercomputers, solving equations that take time to finish. AI models skip the simulation and match patterns in historical weather data instead, which makes them fast. WeatherNext 3 adds a third ingredient: it trains on observations arriving now rather than on archives alone.

“One of the main developments is for [WeatherNext 3] to go beyond what data most global AI models train on. We're able to leverage fresher and richer observational data sets.”

— Samier Merchant, research engineer, Google Research, The Verge, 3 September 2026

Quote source: The Verge, 3 September 2026

The satellite feed matters most where the ground is thin on instruments. Rain gauges are dense across the US and Europe and sparse almost everywhere else, and Google says the biggest gains land in those under-measured regions.

The part that is about electricity

Google built the model to forecast renewable generation as well as weather, including wind speed at 100 metres, roughly the hub height of a turbine. A DeepMind researcher gave the reason directly.

“As the energy needs of Google, but [also] entire humanity, is increasing its energy needs, to make sure that we make renewable a very appealing opportunity is very important for us.”

— Ferran Alet, research scientist, Google DeepMind, The Verge, 3 September 2026

Quote source: The Verge, 3 September 2026

A wind farm that can be predicted an hour ahead at turbine height is worth more to a grid operator than one that cannot, because the operator holds less fossil backup against it. Google is among the largest buyers of that power, and its data centre demand keeps climbing. AI data centres already helped drive a record US clean energy build-out this year.

Google keeps the WeatherNext family documented publicly, and the jump from the previous version is a change of input rather than a bigger network alone: hourly satellite readings instead of six-hourly cycles built on archives. Mining companies and data centre operators buy power on forward contracts, and an error in a wind forecast becomes a line in a settlement bill.

The physics models stay

Weather agencies still run physics models, and WeatherNext 3 trains on their output. Forecasters issue warnings by looking at a spread of predictions rather than one, so an AI model joins the ensemble instead of retiring it. Google says it worked with the US National Hurricane Center and agencies in Asia.

The model now sits inside Search, Maps and Gemini, which is where most people will meet it without noticing.

For anyone running machines that eat electricity, from a hyperscaler to the mining company that swapped its rigs for AI racks this week, an hourly wind forecast at turbine height is a procurement tool. Google shipped a weather upgrade and an input for buying power in the same release.

This piece is informational, not a recommendation to buy, sell, or hold any asset.