Is AI Causing the Climate Crisis — or Helping Solve It?

Artificial intelligence consumes electricity.

It requires expanding data centers, more chips, more cooling systems, more water and more physical infrastructure.

But Google is making the opposite argument as well.

AI, the company says, can also help address climate change.

According to Google’s 2026 Environmental Report, nine of the company’s products and services helped individuals, cities and business partners reduce an estimated 41 million metric tons of carbon dioxide equivalent in 2025.

That figure is nearly three times larger than Google’s own reported ambition-based carbon footprint of about 14.5 million metric tons of carbon dioxide equivalent for the same year.

The implication is powerful.

Google is not only asking whether its own AI infrastructure can become cleaner. It is arguing that AI can help make the wider economy more efficient.

The company’s climate case for AI is concentrated in three areas: energy, transportation and disaster response.

But that case also raises a difficult question.

How should society measure the environmental benefits enabled by AI against the energy, water, materials and emissions required to build and operate the infrastructure behind it?

AI as an Energy-Efficiency Tool

The first part of Google’s argument is energy efficiency.

Google’s Nest thermostats learn household routines and indoor conditions, then adjust heating and cooling settings automatically. The company estimates that Nest devices helped customers save more than 28 billion kilowatt-hours of energy in 2025 and supported roughly 8.8 million metric tons of carbon dioxide equivalent in emissions reductions.

At the level of an individual household, the effect may appear small.

A thermostat lowers heating slightly during the day. It adjusts cooling at night. It reduces unnecessary energy use when residents are away.

But multiplied across millions of homes, those small decisions can affect national electricity demand.

Google’s broader theory is that AI can change the environmental impact of routine decisions at scale.

The same logic applies to solar energy.

Google’s Solar API uses aerial imagery and AI to analyze roof shape, shade from trees and nearby buildings, sunlight exposure and other physical conditions. The system is designed to help solar installers, local governments and energy companies identify buildings where solar installation is likely to be practical and economically attractive.

Google estimates that Solar API supported roughly 1.3 million metric tons of carbon dioxide equivalent in emissions reductions in the United States in 2025.

The technology does not install solar panels itself.

It reduces the uncertainty that can slow solar adoption.

That distinction is important.

Much of AI’s climate value may come not from directly reducing emissions, but from improving the decisions that determine where energy investments are made.

AI and the Transportation System

Transportation is the second major area where Google sees environmental potential.

Google Maps now offers fuel-efficient routes that consider factors such as road grade, traffic conditions, travel speed, vehicle type and expected energy use. Rather than recommending only the fastest route, the system can suggest a route that uses less fuel or electricity while remaining reasonably close in travel time.

Google estimates that fuel-efficient routing supported more than 3 million metric tons of carbon dioxide equivalent in emissions reductions in 2025.

The company also promotes walking, public transit and other lower-carbon transport options when they are competitive with driving.

According to Google, this feature helped shift more than 180 million trips from car travel to lower-carbon transportation options in 2025, supporting an estimated reduction of about 75,000 metric tons of carbon dioxide equivalent.

Traffic management is another AI application.

Google’s Green Light initiative uses Maps data and traffic patterns to help cities improve traffic-signal timing. The goal is to reduce unnecessary stopping, idling and acceleration at intersections.

Google estimates that Green Light supported about 13,000 metric tons of carbon dioxide equivalent in emissions reductions in 2025.

These examples may seem modest compared with the emissions of heavy industry or aviation.

But they point to a broader possibility.

AI can improve the efficiency of systems that already exist.

It can reduce wasted fuel, shorten idle time, guide people toward lower-carbon alternatives and help cities manage infrastructure more effectively.

The value does not come from replacing transportation with software.

It comes from making transportation systems less inefficient.

Aviation, Wildfires and Floods

Google is also applying AI to areas where environmental impacts are harder to see but potentially more significant.

One example is aviation contrails.

Contrails are not simply harmless clouds behind aircraft. Under certain atmospheric conditions, they can trap heat and contribute to aviation’s warming impact.

Google has developed AI tools designed to predict the altitude and weather conditions in which contrails are more likely to form. Airlines may then be able to slightly adjust flight paths or altitude to avoid the conditions that create the greatest warming effect.

Google says its contrail-prediction model generated about 380 metric tons of carbon dioxide equivalent in operational emissions in 2025 while helping support approximately 3,000 metric tons of avoided warming impact.

In other words, Google estimates that the climate benefit was roughly seven times larger than the model’s own operational footprint.

The company is also using AI for disaster response.

By 2025, Google said its flood-forecasting systems were providing river-flood information to more than 2 billion people in around 150 countries. The system is designed to forecast flood risk as much as seven days in advance and distribute alerts through Search, Maps and related warning channels.

Google has also expanded its work on flash-flood forecasting in urban areas, where rapidly changing weather conditions can make disaster response especially difficult.

In wildfire detection, Google is working on the FireSat project, which combines high-resolution infrared satellite imagery with AI analysis to identify fire location, size and intensity more quickly.

The goal is to detect wildfires before they become catastrophic.

That has obvious implications for public safety. But it also matters for climate change because large wildfires can release enormous amounts of carbon into the atmosphere.

Google is pursuing similar work in agriculture.

In India, the company says AI-based monsoon forecasting reached about 38 million farmers, helping them anticipate the beginning of the rainy season and make more informed decisions about planting and crop management.

These services do not eliminate climate risk.

But they may help communities prepare for it.

The Measurement Problem

Google’s argument that AI can enable emissions reductions is plausible.

But the measurement problem remains difficult.

Consider Google Maps.

A fuel-efficient route may save fuel compared with a faster route. But it is hard to know whether every user follows the recommended route, whether traffic conditions develop as expected or whether the route would have been chosen anyway.

The same problem applies to Nest thermostats.

A household may use less energy after installing a smart thermostat. But the reduction may also be influenced by weather, electricity prices, home renovations, changing habits or other efficiency improvements.

Solar installations, traffic-signal upgrades and aviation-routing decisions are even more complicated.

They involve local governments, utilities, airlines, installers, regulators and individual users. Google may provide the data, software or recommendation, but it is not always the only actor responsible for the resulting change.

Google acknowledges some of these limitations.

The company describes its role in different ways depending on the product. In some cases, it may directly reduce emissions through its own operations. In others, it may provide information that supports decisions by users, cities or business partners.

Google also says that its estimates include uncertainty because the company does not always have precise primary data about real-world user behavior and its effects.

That caveat matters.

Corporate claims about “enabled emissions reductions” can be useful because they attempt to quantify the environmental value of digital products.

But they should not be read as the same thing as direct, independently verified emissions reductions.

They are estimates of what may have been avoided because people, cities or companies used a particular tool.

That is a meaningful category.

But it is not a simple one.

AI’s Benefits Cannot Automatically Offset Its Infrastructure Costs

The larger concern is that AI-enabled climate benefits could be used to justify unlimited AI infrastructure expansion.

AI may optimize traffic.

It may reduce household energy waste.

It may improve solar adoption.

It may help forecast floods and detect wildfires.

But AI also requires data centers, electricity, cooling systems, semiconductors, server manufacturing and construction.

The infrastructure required to run AI has its own environmental footprint.

That means AI’s climate value cannot be assessed only by counting the emissions it helps avoid elsewhere.

It must also be assessed by considering the emissions, water use and resource consumption generated by the systems that make those services possible.

Google itself recognizes this tension.

The company says it is trying to maximize AI’s environmental benefits while minimizing the resource intensity of the AI infrastructure behind it.

That is the right framing.

AI cannot be considered climate-positive simply because it supports one useful environmental application. Its net value depends on whether the benefits it enables exceed the environmental costs required to build and operate it.

A flood-prediction model may save lives and reduce damage.

A traffic system may reduce unnecessary idling.

A smart thermostat may lower household energy demand.

But if the data centers supporting those services rely heavily on fossil-fuel electricity or place severe pressure on water-stressed communities, the net environmental benefit becomes harder to evaluate.

The Two Questions That Define Sustainable AI

The climate value of AI should therefore be judged through two questions.

First, how much does AI actually reduce emissions, environmental damage or climate risk across society?

Second, how much electricity, water, materials and carbon are required to build and operate the AI systems themselves?

A good AI system is not simply one that produces accurate results.

It is one that creates the greatest environmental benefit with the least possible resource use.

Google wants to position AI as a climate solution rather than a climate problem.

That ambition is not unreasonable.

AI can help people make better choices, help cities run more efficiently and help communities respond to disasters before they become more destructive.

But the claim will become convincing only if companies disclose both sides of the equation.

They must show not only the emissions their AI services may help avoid, but also the emissions, energy demand and resource consumption created by the infrastructure behind those services.

AI could become part of the climate solution.

But that outcome will not depend only on how intelligent the technology becomes.

It will depend on how much change AI can create with how little of the planet it consumes.