What happened
Google's WeatherNext 3 page, updated September 3 in the lab's sitemap, positions the model as an operational data product rather than a research demo. It generates forecasts every hour of the day, which is a meaningful change for fast-moving local conditions. Earlier global models commonly ran on fixed schedules, leaving longer gaps between updates.
The model draws directly from raw satellite imagery. Google says that input design lets WeatherNext 3 update frequently and produce local predictions even in places it did not encounter during training. The page specifies five- kilometer resolution for weather-station-targeted temperature and humidity, and ten-kilometer resolution for other surface variables including wind.
WeatherNext 3 is an ensemble model, meaning it generates a distribution of possible outcomes rather than one deterministic future. That is useful for decisions where uncertainty matters: a grid operator does not just want the most likely wind speed, but a sense of the range that could affect generation. Google calls out radiation and cloud-cover variables for wind and solar operators as an intended enterprise use.
Distribution is unusually broad for a scientific model. Google says it is being integrated into Search, Maps, and Gemini, giving consumers local weather information through products they already use. Enterprise customers can access real-time operational data and historical forecasts in BigQuery, Earth Engine, Google Maps Platform, and Google Cloud Storage without running the forecast model themselves.
The release builds on WeatherNext work in tropical cyclones. An August research update described improvements in cyclone forecasting. Google also provides Weather Lab, an experimental interface for comparing AI models with traditional meteorological baselines and tracking tropical cyclones.
Why it matters
Weather is a strong example of AI moving from a chatbot into infrastructure. The output can change when a delivery route departs, how much renewable power a grid schedules, or whether an outdoor crew works. Value comes from timeliness, geographic detail, and calibrated uncertainty, not conversational style.
Hourly global forecasts could be especially useful where observations are sparse or conditions change quickly. Direct satellite input reduces reliance on a dense local station network, although local observations remain critical for verification. Integrating the results into common Google products also shortens the path between a forecast improvement and a user decision.
For enterprises, the data interfaces matter as much as the model. A forecast locked inside a paper is difficult to operationalize. Historical and real-time records in familiar cloud systems make it possible to test whether the model improves a specific decision before changing production operations.
The fine print
Google says Weather Lab is experimental and directs people to local meteorological agencies or national weather services for official forecasts and warnings. That distinction should remain visible wherever an application uses the data. Resolution is not the same as certainty, and a model's average accuracy may conceal important failure modes during rare, high-impact events.
The forecast now refreshes every hour. The weather has agreed to continue ignoring the release schedule.