Google is moving its newest AI weather model into the operational market for renewable energy. WeatherNext 3 produces hourly global forecasts at resolutions down to five kilometres and includes variables directly relevant to electricity production, such as wind speed at 100 metres, cloud cover and sunlight reaching the surface.
Google DeepMind and Google Research released the model on September 3. It now feeds consumer weather results across Google services, but its more consequential commercial role may sit behind those interfaces: grid operators, energy traders and renewable developers can query the data through BigQuery and Earth Engine or download it from Google Cloud Storage.
Hourly forecasts target a harder balancing problem
WeatherNext 3 improves on WeatherNext 2’s 25-kilometre grid and six-hour refresh cycle. Google says the higher spatial detail and hourly updates can help operators estimate output from wind and solar assets, then match variable generation against demand.
The need grows as both sides of the grid become less predictable. S&P Global Market Intelligence projects 51.2 gigawatts of solar and 25.7 gigawatts of energy storage among more than 90 gigawatts of planned US capacity additions in 2026. At the same time, S&P identifies expanding North American data centres as a major source of electricity-demand growth.
Deloitte projects peak demand rising about 26% by 2035, with data centres potentially reaching 176 gigawatts, five times their 2024 level. In that setting, a forecast error can have immediate costs. Underestimating wind may force an operator to procure replacement power at short notice; overestimating it can leave renewable generation curtailed because the grid cannot absorb the expected supply.
Distribution gives Google a commercial advantage
Specialist vendors already sell energy-focused weather products. The source names Vaisala, Solcast, DNV’s WindGEMINI, IBM’s HyperWatch and Swiss provider Jua. Google enters that market with an unusually broad delivery network: the same forecast can appear in Search and Maps, as a BigQuery table, as an Earth Engine layer or through a platform API.
That reach reduces the setup needed to start using the model and closes an update-frequency gap that specialist competitors have emphasized. It does not settle the accuracy question. Jua argues that physics-based models can outperform purely data-driven systems during record-breaking extreme weather, because physical rules remain useful when conditions fall outside learned patterns. Jua sells a physics-constrained product, so its claim also reflects a commercial interest.
Google describes a technical change in how WeatherNext 3 handles observations. The model ingests live geostationary satellite imagery and trains directly on individual weather-station readings, reducing reliance on numerical weather prediction outputs that arrive with a lag. Google’s own architecture still includes traditional historical analysis, and reporting places the remaining delay at three to four hours rather than eliminating it.
Performance claims still require local proof
Google reports forecast improvements of up to 60% against NASA’s IMERG satellite product, 30% against MRMS radar and 10% against rain-gauge readings at early lead times. A separate claim of 50% better precipitation forecasting applies to forecasts at least a day ahead. These percentages use different baselines and all contain an “up to” qualification, so they cannot be added together or treated as typical performance.
No independent third-party validation accompanied the launch. Google points to live evaluations by Brightband when calling WeatherNext 3 the most accurate global model to date, but an electricity operator will need evidence for its own territory, assets and extreme-weather exposure.
Commercial terms are another open question. Google has not disclosed enterprise pricing or clarified whether BigQuery and Earth Engine access will use standard cloud-query charges or a separate licence. The company also has its own stake in better renewable forecasting: hyperscale data centres are contributing to load growth, while Google has signed large renewable procurement agreements for its facilities.
WeatherNext 3 therefore arrives with scale, frequent updates and convenient distribution, but energy customers can test it against concrete operational measures. Local forecast accuracy, resilience during unusual conditions and total access cost will determine whether Google’s reach becomes an advantage on the grid.
