Google has been quietly building toward this moment for a while. Its AI-based weather models have been turning heads in research circles — outperforming traditional physics-based forecasting in both accuracy and speed — and now the company is ready to move them out of the experimental phase and into real products.
“We’re taking it out of the lab and really putting it into the hands of users in more ways than we have before,” remarked Peter Battaglia, senior director of research and sustainability at Google DeepMind, in a briefing with reporters. “We have confidence that our forecasts are effective and useful.”
The model driving this shift is WeatherNext 2. It generates forecasts eight times faster than its predecessor and hits an accuracy rate of 99.9 percent for variables like temperature and wind. What makes that especially striking is what it can do in under a minute: produce hundreds of possible outcomes from a single starting point — something that would take traditional supercomputer models several hours to accomplish.
Standard weather forecasting has always leaned hard on complex computational methods designed to simulate atmospheric physics. AI models take a different route entirely, scanning historical weather data for patterns and using those to project future conditions.
To make WeatherNext 2 work at this scale, Google built in a new technique called the Functional Generative Network (FGN). Older models had to run repeated calculations for every forecast. FGN folds controlled randomness directly into its inputs, so it can produce a wide range of outcomes in one pass rather than many.
The practical result is a model that can forecast weather up to 15 days out and deliver hourly predictions. Google expects both enterprise clients and everyday users will find that useful.
“We’ve seen significant interest from sectors such as energy, agriculture, transportation, and logistics,” shared Akib Uddin, a product manager at Google Research. “These one-hour forecasts help businesses make informed decisions that affect their operations.”
WeatherNext 2 is being woven into platforms including Maps, Search, Gemini, and Pixel Weather. Google is also launching an early access program for businesses that want customized modeling. The underlying forecast data will be available through Google Earth Engine for geospatial work and BigQuery for large-scale data management — a sign that the company sees AI-driven forecasting as something with serious practical legs, not just a flashy demo.
