Google’s WeatherNext 3 Uses AI to Predict Wind and Solar Power Every Hour

 

Google's newest artificial intelligence weather model is not just trying to tell you whether you need an umbrella. It could help decide how much electricity a wind farm or solar plant will generate in the next few hours.

Google DeepMind and Google Research have launched WeatherNext 3, their most advanced global AI weather forecasting system to date.

Released on September 3, 2026, the model produces a new global forecast every hour, uses live satellite observations, and offers significantly higher spatial resolution than its predecessor.

But one of the most important additions may be aimed not at ordinary weather users, but at the energy industry.

WeatherNext 3 can forecast wind speeds around 100 meters above the ground, roughly the operating height of many modern wind turbines. It can also predict cloud cover and solar radiation reaching the Earth's surface.

Those variables can help wind and solar operators estimate how much electricity their facilities are likely to produce.

And that makes WeatherNext 3 much more than another weather app.

It puts Google deeper into the business of energy forecasting, grid management, renewable power and enterprise weather intelligence.

From Six-Hour Forecasts to Every Hour

The technical improvement over Google's previous generation is significant.

WeatherNext 2 produced forecasts on a roughly 25-kilometer grid with six-hour intervals.

WeatherNext 3 now generates forecasts every hour.

For some surface variables, including targeted temperature and humidity forecasts, Google can provide resolution as fine as five kilometers.

Other surface variables can operate at approximately 10-kilometer resolution, while some atmospheric variables are produced at around 25 kilometers.

So it would be inaccurate to say that every WeatherNext 3 forecast operates at five-kilometer resolution.

What Google has achieved is a much more detailed and frequently updated global forecasting system that can adapt to different variables and applications.

Why Wind Farms Care About Weather at 100 Meters

Weather near the ground is not necessarily the same as weather where turbine blades are spinning.

That is why WeatherNext 3 specifically predicts wind conditions at around 100 meters above the surface.

For a wind farm operator, knowing expected wind speed directly affects predictions of how much electricity turbines can produce.

For solar operators, cloud cover and solar radiation matter just as much.

A clear sky can mean strong solar production.

Unexpected cloud cover can sharply reduce generation.

Google says WeatherNext 3 was specifically designed to forecast these renewable-energy variables so operators can estimate clean-energy output and better match generation with electricity demand.

This is where weather forecasting becomes an economic tool.

The question is no longer simply:

Will it be windy tomorrow?

The question becomes:

How many megawatts will that wind produce at 2:00 p.m.?

And:

How much backup electricity will the grid need if that forecast changes?

Renewable Energy Makes Accurate Forecasting More Valuable

Electricity grids are becoming increasingly dependent on weather.

According to S&P Global Market Intelligence, the United States is projected to add more than 90 gigawatts of new electricity capacity in 2026.

Around 51.2 gigawatts is expected to come from solar, while approximately 25.7 gigawatts will come from energy storage. Wind is projected to add another 13.1 gigawatts.

Solar and wind are different from traditional power plants.

A gas turbine can generally produce electricity when an operator tells it to.

A solar farm cannot tell the sun to shine.

A wind turbine cannot order the wind to blow.

That means grid operators need increasingly accurate forecasts to know how much renewable electricity will be available several hours or days ahead.

If the forecast is wrong, the consequences can become expensive.

If operators expect more wind power than actually arrives, they may need to purchase electricity at short notice or activate more expensive backup generation.

If renewable production exceeds what the grid can absorb, operators may have to curtail generation.

Better forecasts can therefore translate directly into better grid economics.

The Irony: AI Is Also Creating More Electricity Demand

There is another side to the story.

Artificial intelligence is helping utilities manage electricity demand.

But artificial intelligence is also contributing to the enormous growth in electricity consumption that utilities are struggling to serve.

AI data centers require massive amounts of power.

Deloitte projects that US peak electricity demand could increase by approximately 26 percent by 2035.

Data-center electricity demand alone could reach about 176 gigawatts by 2035, several times current levels.

In other words:

AI is increasing pressure on the grid while simultaneously becoming one of the technologies being used to manage that pressure.

The same hyperscale technology industry building enormous AI computing facilities is also developing forecasting, optimization and energy-management systems designed to make the grid more intelligent.

WeatherNext 3 Is Already Inside Google Products

Most consumers may encounter WeatherNext 3 without even knowing its name.

Google says the model is being integrated into:

Google Search

Gemini

Google Maps

Google Maps Platform Weather API

Google Earth Engine

and other Google Cloud services.

But the enterprise layer may prove commercially more important.

Businesses and researchers can access WeatherNext 3 forecast data through BigQuery and Earth Engine or download datasets from Google Cloud Storage.

That means a renewable-energy company does not necessarily need to install and operate the forecasting model itself.

It can consume the forecast as data and combine it with its own systems.

A wind-energy company could combine WeatherNext forecasts with turbine information.

A solar operator could combine the forecasts with historical generation data.

An electricity trader could integrate weather projections into pricing models.

A grid operator could use them alongside electricity-demand forecasts.

That makes WeatherNext 3 potentially useful as an underlying data layer rather than simply a standalone weather product.

Google Is Entering an Existing Weather Intelligence Market

Google is not inventing commercial weather forecasting.

Energy companies already pay specialist providers for sophisticated forecasting and risk information.

The difference is Google's scale.

Google can distribute WeatherNext 3 through its existing ecosystem of Search, Maps, Gemini, Earth Engine, BigQuery and Google Cloud.

Few specialist weather companies can match that distribution network.

This could allow weather intelligence to become another layer of Google's cloud and AI infrastructure.

A company already storing operational information inside Google Cloud could potentially combine its own business data with WeatherNext forecasts without creating an entirely separate weather-data architecture.

That could make AI weather forecasting considerably easier to deploy.

WeatherNext 3 Learns More Directly From Real Observations

One of the most interesting technical changes involves the data feeding the model.

Many previous AI weather systems relied heavily on information generated by numerical weather prediction, or NWP.

These are enormous physics-based forecasting systems operated using supercomputers.

Google says conventional analysis data can carry significant latency, which can become particularly important when weather conditions are changing rapidly.

WeatherNext 3 introduces direct ingestion of hourly geostationary satellite mosaics.

It also trains directly on observations from individual weather stations for certain surface variables.

The result is a system that can incorporate much more recent information and produce forecasts every hour.

But there is an important distinction.

WeatherNext 3 has not completely abandoned traditional weather-analysis data.

Google's own architecture shows the system combining live satellite mosaics with historical analysis information.

So the breakthrough is better described as reducing dependence on delayed model-derived information and adding much more direct observational data, rather than completely replacing physics-based weather forecasting.

Google Claims Major Precipitation Improvements

Google says WeatherNext 3 also improves precipitation forecasting.

For forecasts made a day or more in advance, Google reports improvements of up to 50 percent in precipitation accuracy compared with previous capabilities.

The company also reports different gains against satellite, radar and rain-gauge benchmarks depending on forecast lead time and evaluation method.

Those numbers require some caution.

"Up to" means the strongest measured result, not necessarily the average improvement everywhere.

Weather performance also varies dramatically by:

location,

weather type,

forecast horizon,

terrain,

season,

and measurement method.

A utility deciding whether to rely on an AI forecast would therefore care much more about performance around its own wind farms, solar facilities or service territory than about one global average.

Google points to independent live evaluations from Brightband's Operational WeatherBench in support of its accuracy claims.

Brightband's benchmark currently evaluates WeatherNext 3 alongside traditional and AI systems including ECMWF models, NOAA forecasts, GraphCast, WeatherNext 2, Microsoft Aurora and NVIDIA Atlas.

That kind of continuing real-world evaluation will matter as AI weather systems move from research demonstrations into operational infrastructure.

AI Weather Forecasting Is Becoming Infrastructure

This is probably the most important part of the story.

Weather AI used to be mostly framed as a scientific competition:

Can an AI model predict weather better than conventional models?

The question is changing.

Now businesses are asking:

Can we use these forecasts to make better operational decisions?

Airlines can optimize flights.

Farmers can plan irrigation and harvesting.

Shipping companies can adjust routes.

Insurance companies can model risk.

Solar companies can predict production.

Wind farms can estimate turbine output.

Electric utilities can balance supply and demand.

At that point, weather forecasting stops being simply weather information.

It becomes part of the digital infrastructure of the economy.

What This Means for the Philippines

The Philippines should pay close attention to this development.

We are one of the countries where weather influences almost every part of the economy.

Typhoons affect electricity systems.

Cloud cover affects solar production.

Wind conditions affect wind farms.

Heavy rainfall affects hydroelectric generation.

Weather disrupts agriculture, airports, ports, shipping, logistics and telecommunications.

And because the Philippines is an archipelago, localized weather differences can be enormous.

Conditions in Ilocos may be very different from Bicol.

Weather in Metro Manila may have little resemblance to Mindanao.

Rainfall on one side of a mountain range can differ dramatically from conditions only a short distance away.

That makes higher-resolution, frequently updated forecasting particularly valuable.

Imagine Philippine renewable-energy operators combining hourly AI forecasts with actual generation data from:

Bangui and Burgos wind farms,

solar farms across Luzon and the Visayas,

hydroelectric facilities,

and eventually offshore renewable projects.

An intelligent grid could continuously compare expected renewable generation with electricity demand.

If solar production is expected to fall because of cloud cover, other generation or battery capacity could be prepared earlier.

If wind generation is expected to rise, storage systems could be scheduled differently.

That could eventually help reduce waste, improve reliability and support a Philippine grid that increasingly integrates renewable energy.

But one rule must remain clear.

AI-generated forecasts should not replace official severe-weather warnings.

Google itself says WeatherNext is not intended to replace national meteorological agencies for official alerts and safety information. In the Philippines, people should continue to rely on PAGASA and authorized government agencies for official weather warnings and disaster advisories.

WeatherNext 3 therefore represents something bigger than a better forecast inside Google Search.

Google is turning weather prediction into another AI platform.

And the companies that can predict the wind, the clouds, the rain and the sunlight more accurately may eventually influence something even more important:

where the world's electricity comes from, how much of it will be available, and what it will cost.

The AI race is no longer only about who builds the smartest chatbot.

It is increasingly about who can use artificial intelligence to understand and manage the physical world itself.