How Much Energy Does AI Use? Prompts, Data Centers and CO2

Hakkı ALKAN

October 31, 2026

How much energy does AI use? A single text prompt is small: Google measured 0.24 watt-hours for the median Gemini Apps prompt, and OpenAI CEO Sam Altman put the average ChatGPT query at about 0.34 watt-hours. The real cost shows up at scale. The International Energy Agency (IEA) expects data centers worldwide to use around 945 TWh of electricity by 2030, and a new Sierra Club analysis says US power plant emissions could rise 42% by 2030 as AI data centers come online.

Here is what those numbers mean, where they come from and why the estimates vary so much.

How Much Energy Does AI Use per Prompt?

The two most cited per-prompt figures come from the companies themselves:

• Google Gemini: In August 2025, Google published a technical paper estimating that the median Gemini Apps text prompt (measured in May 2025) used 0.24 Wh of energy, emitted 0.03 grams of CO2 equivalent and consumed 0.26 milliliters of water, about five drops. Google compares the energy to watching TV for less than nine seconds.
• OpenAI ChatGPT: In his June 2025 blog post “The Gentle Singularity,” Sam Altman wrote that the average ChatGPT query uses about 0.34 Wh, roughly what an oven uses in a little over one second, plus about 0.000085 gallons of water.

These figures are useful, but they have limits. Google’s number is a median for text prompts only, and the company notes it has not been verified by an independent third party. OpenAI has not published a methodology for its figure. Neither includes the energy used to train the models or to manufacture the servers.

Why do some prompts use much more energy?

Energy use depends on how much work the model does. A short question with a short answer is cheap. Long documents, reasoning models that “think” through many intermediate steps before answering, and image or video generation all require far more computation, so the per-request energy can be many times higher than the headline figures.

Measurement choices matter too. Google says that if it counted only the active AI chips, the median prompt would use just 0.10 Wh. Its 0.24 Wh figure also includes host CPUs and memory, idle machines kept ready for traffic spikes, and data center overhead such as cooling.

How Much Electricity Do AI Data Centers Use in Total?

Small numbers per prompt add up when billions of requests run every day on huge clusters of GPUs and custom chips. The best public estimates look like this:

• Worldwide: The IEA’s Energy and AI report estimates data centers used about 415 TWh in 2024, around 1.5% of global electricity. In its base case, that roughly doubles to about 945 TWh by 2030, slightly more than Japan’s total electricity use today and just under 3% of global demand. Electricity use by AI-focused accelerated servers grows about 30% a year in that scenario.
• United States: A Department of Energy-backed Lawrence Berkeley National Laboratory report found US data centers used 176 TWh in 2023, or 4.4% of the country’s electricity, up from 58 TWh in 2014. It projects 325 to 580 TWh by 2028, equal to 6.7% to 12% of US consumption.

The US is the biggest driver. According to the IEA, data centers account for nearly half of US electricity demand growth through 2030. The same hardware boom is also pushing up memory prices, as we explained in why RAM is so expensive right now.

What Does AI Energy Use Mean for Carbon Emissions?

The climate impact depends less on the chips than on what powers them. That is where the latest warning comes in. An internal Sierra Club analysis, reported by The Verge and Bloomberg this week, projects that the US power sector would emit 2.1 billion metric tons of CO2 in 2030 if planned AI projects are built. That is 618 million tons more than in 2025, an increase of about 42%, which would erase roughly 15 years of emission cuts from US electricity generation.

Most of that increase would come from off-grid “bring your own power” plants, usually gas turbines that data center developers build to avoid long waits for a grid connection. Jeremy Fisher, the Sierra Club’s principal advisor for climate and energy, estimates those sources alone could exceed 460 million tons a year by 2030, close to Mexico’s annual emissions. Grid-connected generators would add another 154 million tons. AI companies are already making this choice: Anthropic, for example, has turned to natural gas to power AI data centers.

The Sierra Club’s figure is at the high end. The Union of Concerned Scientists projected in January that fossil-fueled data center power could raise US power-sector emissions by 19% to 29% by 2035, and a study by US and Canadian researchers published in June projected that data centers and crypto mining could push power-sector emissions 28% higher by 2030 than in a scenario without that growth. The spread comes from open questions: how many announced data centers actually get built, how efficient their hardware is and how much clean energy comes online in time.

Is Using AI Bad for the Environment?

For an individual user, a few dozen text prompts a day is a small share of personal energy use, based on the companies’ own figures. The bigger issue is systemic: total demand is growing fast, and in the US much of the new supply is natural gas. Tech companies are responding with efficiency gains (Google says the energy per median Gemini prompt fell 33 times between May 2024 and May 2025), clean energy contracts, nuclear deals and even experiments such as Google’s Project Suncatcher, which tests AI computing on solar-powered satellites.

If you want to cut your own AI footprint, the simple rules apply: use smaller or faster models for simple tasks, avoid regenerating answers needlessly and save heavy image or video generation for when you really need it.

Frequently Asked Questions

How much energy does one ChatGPT question use?

OpenAI says an average ChatGPT query uses about 0.34 watt-hours. It has not published how it measured that figure, and long or reasoning-heavy requests use more.

How much water does AI use per prompt?

Google estimates 0.26 milliliters for a median Gemini text prompt, and OpenAI says about 0.000085 gallons (roughly 0.32 milliliters) for an average ChatGPT query. Both figures cover data center operation, not the water used to generate the electricity or build the hardware.

What share of electricity do data centers use?

About 1.5% worldwide in 2024, according to the IEA, and 4.4% in the US in 2023, according to Lawrence Berkeley National Laboratory. The US share could reach 6.7% to 12% by 2028.