What happened
Google DeepMind released WeatherNext 3, an AI weather forecasting model specifically designed for wind farms, solar producers, and electricity grid operators. The model updates every hour and generates two primary forecast streams: wind speed and direction at heights up to 100 meters, and cloud cover with incoming solar radiation. This hourly resolution is intended to help operators anticipate fluctuations in renewable energy output and manage grid stability more effectively.
Google DeepMind announced the launch of WeatherNext 3 on September 14, describing it as a tool for energy infrastructure operators. The model is built to forecast conditions that directly impact renewable power generation, specifically targeting wind and solar sectors.
The system operates on an hourly update cadence, which DeepMind states is essential because weather conditions affecting power output can shift within a single trading period on power markets. The model produces two distinct data streams: one for wind speed and direction at turbine-relevant heights (up to 100 meters) and another for cloud cover and solar radiation to estimate photovoltaic generation.
DeepMind positioned this release as infrastructure for the energy transition, arguing that advance notice of changing weather allows teams to balance power grids when renewable generation is variable. This builds on the company's earlier work with GraphCast, a global weather model that previously outperformed traditional benchmarks.
Why it matters
WeatherNext 3 addresses a critical challenge in the energy transition: the variability of wind and solar power. By providing high-resolution, hourly forecasts, the model allows grid operators to better match clean energy supply with consumer demand, reducing the need for costly backup generation and improving the efficiency of power markets. This represents a practical application of AI in critical infrastructure, moving beyond general weather prediction to specialized utility management.
Electricity grids have historically relied on dispatchable generation, such as coal, gas, and nuclear plants, which can be throttled on demand. Wind and solar are non-dispatchable, meaning their output depends on uncontrollable weather conditions. As the share of renewables grows, grids must anticipate generation fluctuations rather than react to them.
Forecast errors in renewable energy can have compounding effects on grid stability. Overestimating wind output can lead to supply shortages during peak demand, while underestimating it wastes capacity and forces the use of costlier backup generation. Hourly-resolution AI forecasting aims to mitigate these risks by providing accurate data for storage dispatch and demand response.
The deployment of such a specialized AI model by a leading lab like DeepMind raises the competitive floor for forecast accuracy in the sector. It also highlights the growing intersection of AI and critical infrastructure, where precise data is essential for managing the complexity of a decarbonizing power grid.
What to watch next
The next steps involve observing which specific grid operators or energy companies adopt WeatherNext 3, as DeepMind has not disclosed current users. Additionally, the model's accuracy over extended forecast horizons, particularly beyond six hours, will be crucial for its utility in day-ahead power market bidding, where forecast errors carry significant financial penalties.
DeepMind has not disclosed which grid operators, utilities, or independent power producers are currently using WeatherNext 3. Monitoring adoption by major energy companies will indicate the model's practical impact and commercial viability.
The model's performance in day-ahead power market bidding is a key metric to watch. Since the largest volume of electricity contracts settles in this window, the accuracy of forecasts beyond the six-hour mark will determine its financial utility for energy traders and operators.
Competitors, including startups like Tomorrow.io and national meteorological agencies, are also targeting this market. The entry of DeepMind may accelerate innovation and pressure existing services to improve their forecast accuracy and resolution.