Google DeepMind and Google Research dropped WeatherNext 3 on September 3, and it’s not just another consumer weather app upgrade. This model predicts wind speed at 100 meters—the exact height of modern turbine blades—along with cloud cover and surface sunlight. It refreshes every hour at a five-kilometer resolution, a massive leap from the six-hour, 25-kilometer grid of its predecessor. While casual users will see smoother forecasts in Google Maps and Search, the real money is in the enterprise layer. Energy traders and grid operators can now pull this data directly from BigQuery or Earth Engine without setting up complex models.
The timing isn’t accidental. The energy grid is facing a perfect storm of unpredictability. On one side, renewables are dominating new capacity, with S&P Global projecting 51.2GW of solar and 25.7GW of storage additions this year alone. These sources depend entirely on weather conditions. On the other side, AI data centers are driving electricity demand through the roof. Deloitte estimates peak demand could jump 26% by 2035, with data center usage hitting 176GW. When forecasts fail, the costs are immediate. Underestimate wind power, and you buy expensive gas backup. Overestimate it, and you pay renewable farms to shut down because the grid can’t handle the surplus.
Google is stepping into a crowded ring. Incumbents like Vaisala, IBM’s HyperWatch, and Swiss startup Jua already sell specialized forecasts. Jua argues that physics-based models still beat pure AI during extreme weather events because they understand atmospheric rules rather than just pattern-matching past data. But Google’s trump card is distribution. No other vendor offers the same forecast across Search, APIs, and cloud databases simultaneously. By training on live satellite imagery and station readings instead of lagged simulations, WeatherNext 3 cuts out the six-hour delay inherent in traditional numerical models.
This shift from simulation to observation matters for fast-changing variables like rain and temperature. Ilan Price, a senior research scientist at DeepMind, noted the gain comes from using the most recent information available rather than waiting for the next analysis cycle. While competitors may hold an edge in rare, record-breaking storms, Google’s hourly refresh rate closes the frequency gap that specialists have relied on. For grid operators balancing AI’s hungry servers against fickle wind farms, having fresher data might be worth more than theoretical perfection in extreme edge cases.
