Do You Know How Much Electricity One ChatGPT Question Actually Uses?

Quick answer: OpenAI says the average ChatGPT question uses about 0.34 watt-hours of electricity, roughly what a kitchen oven draws in a little over one second, and about 0.000085 gallons of water, close to one-fifteenth of a teaspoon. Independent research from Epoch AI puts a typical GPT-4o query even lower, near 0.3 watt-hours, about ten times less than the 3-watt-hour figure that circulated online for years. Multiply either number by the roughly 2.5 billion prompts ChatGPT handles in a day and the electricity use of one tiny question turns into a very large number. Here is what the real data says, where the numbers come from, and where they are still shaky.

I cover AI tools and their real-world footprint for a living, and few questions come up more often in the comments on my Instagram and Facebook AI posts than this one: does typing a question into ChatGPT or Gemini actually cost meaningful electricity? The honest answer is more interesting than the viral claims. Let’s look at what OpenAI, Google, and independent researchers have actually published, and do the math nobody else bothers to finish.

The Real Number: What OpenAI Says One ChatGPT Question Costs

In a blog post published on June 11, 2025, OpenAI CEO Sam Altman gave the company’s own figure for the first time. In his words, “the average query uses about 0.34 watt-hours, about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes.” He added that it uses about 0.000085 gallons of water, “roughly one-fifteenth of a teaspoon.” (Source: Data Center Dynamics)

Those figures came with real limits. Altman did not publish the methodology behind them, and energy researchers who reviewed the claim pointed out that OpenAI never defined what an “average query” includes, whether longer or more complex requests were folded into the average, and whether the number accounts for the wide range of hardware and cooling setups across Microsoft’s data centers. The figure also only covers running the model (inference), not the far larger amount of electricity spent training it in the first place.

Where the “ChatGPT Drains a Bottle of Water Every Time” Myth Came From

For years, the most-cited number for a ChatGPT query was 3 watt-hours, about ten times higher than what OpenAI itself now claims. That figure traces back to an early, widely shared estimate that was never confirmed by OpenAI and was based on assumptions about older, less efficient hardware.

Epoch AI, an independent nonprofit AI research organization, ran its own numbers in a report published February 7, 2025. Using OpenAI’s GPT-4o model, roughly 100 billion active parameters, a standard 500-token response, and Nvidia H100 GPUs running at around 10 percent utilization, Epoch AI calculated a typical query at approximately 0.3 watt-hours, in line with Altman’s own later figure and about ten times lower than the old viral number. For context, Epoch AI notes that 0.3 watt-hours is “less than the amount of electricity that an LED lightbulb or a laptop consumes in a few minutes,” and that a single 2009-era Google search was estimated at about the same 0.3 watt-hours. (Source: Epoch AI)

Not every query is that cheap, though. Epoch AI’s model shows the cost climbs with input size and reasoning: a query carrying 10,000 input tokens runs closer to 2.5 watt-hours, a query using the maximum 100,000-token context window can reach roughly 40 watt-hours, and reasoning-focused models that “think” through extra steps before answering can generate about 2.5 times more output tokens than a standard response, pushing energy use higher still.

How Google’s Gemini Compares

Google published its own per-prompt figures in a technical paper in August 2026. According to that paper, a single Gemini text prompt uses about 0.24 watt-hours of electricity, produces roughly 0.03 grams of CO2-equivalent emissions, and consumes about 0.26 milliliters of water, close to five drops. Google described this as “substantially lower than many public estimates” and compared it to running a television for under nine seconds. (Source: Google technical paper, arXiv)

Google’s own disclosure comes with the same caveat as OpenAI’s: the number covers only inference for standard text prompts in the Gemini app, not training, not image or video generation, and not the heavier “deep research” style tasks. Google itself noted the data “hasn’t been independently verified and could change as new models are added,” and separately reported that its total company-wide emissions rose 51 percent over five years even as per-prompt efficiency improved, a reminder that per-query savings do not automatically shrink the total footprint once billions of queries are added up.

What Billions of Queries a Day Actually Add Up To

A single query barely registers. Billions of them do not. OpenAI has reported ChatGPT handling roughly 2.5 billion prompts a day (a figure OpenAI gave as of July 2025) and reaching 900 million weekly active users by February 2026. Since ChatGPT’s user base kept growing between those two data points, today’s actual daily query volume is very likely higher than 2.5 billion, which makes the numbers below a conservative floor rather than an overestimate.

Here is NexPloreInfo’s own calculation, run on the publicly disclosed figures above, using Epoch AI’s household baseline of about 28,000 watt-hours (28 kWh) of electricity used by an average U.S. home per day:

In other words, on OpenAI’s own math, the text questions typed into ChatGPT in a single day use roughly as much electricity as 30,000 average American homes use in that same day, and that is before counting image generation, video generation, “deep research” tasks, or any of Google’s, Anthropic’s, or xAI’s competing chatbots. Nvidia’s AI chips, the hardware powering most of this demand, are exactly why the company became the world’s first $5 trillion business.

The Bigger Picture: A 2030 Forecast, Not a Current Fact

Zooming out from individual chatbots to the data centers that run them, the International Energy Agency’s Electricity 2026 and Energy and AI reports give the clearest independent numbers available. Data centers used around 415 terawatt-hours of electricity worldwide in 2024, about 1.5 percent of total global electricity consumption. The IEA forecasts, and this is a projection, not a confirmed current figure, that data center electricity use will more than double to around 945 terawatt-hours by 2030, a total that would exceed Japan’s entire current electricity consumption. The IEA names AI as “the most important driver” of that growth, and projects the United States alone will account for roughly half of the worldwide increase, with U.S. data centers on track to consume more electricity by 2030 than the country’s aluminium, steel, cement, and chemical production combined. (Source: International Energy Agency)

That forecast lines up with what is already visible on the ground. Microsoft alone has announced plans to triple its data center capacity to 38 gigawatts by 2032, and every major AI lab, including Anthropic, is racing to build out similar infrastructure to keep up with demand.

Why These Numbers Are Still Disputed

Three honest caveats belong here, because a “Do You Know” fact is only useful if it is not oversimplified. First, both OpenAI’s and Google’s per-query figures are self-reported, not audited by an outside body, and Google has said its own numbers “could change.” Second, none of the per-query figures include the electricity spent training the underlying models in the first place, which independent researchers generally agree is a far larger one-time cost than any number of individual queries. Third, “average query” is doing a lot of work in these numbers. A one-line question and a request asking the model to analyze a 50-page document are not remotely the same electricity cost, even though both get folded into the same published average.

What This Actually Means for You

None of this means a single ChatGPT or Gemini question should keep anyone up at night. On OpenAI’s and Google’s own numbers, one question uses less electricity than boiling a kettle, running a hair dryer for a few seconds, or leaving a phone charger plugged in overnight. The environmental conversation that actually matters is happening at the scale of billions of queries and the power plants being built to run them, not at the scale of any one person’s habits. If you use AI tools daily, as I do, the more useful thing to watch is not how you ask a question but what happens to your data once you have asked it, and whether the platform you are using is transparent about both.

Frequently Asked Questions

How much electricity does one ChatGPT question use?
OpenAI says about 0.34 watt-hours per average query, roughly what a kitchen oven uses in a little over one second. Independent research from Epoch AI estimates a similar figure, around 0.3 watt-hours.

Is the “ChatGPT uses a bottle of water per question” claim true?
No. OpenAI’s own figure is about 0.000085 gallons of water per query, roughly one-fifteenth of a teaspoon, far below a bottle. That said, this covers only running the model, not the water used to manufacture the hardware or cool the wider data center campus.

Does Google’s Gemini use less energy than ChatGPT?
Based on each company’s own published figures, yes: Google reports about 0.24 watt-hours per Gemini text prompt versus OpenAI’s 0.34 watt-hours for ChatGPT. Both figures are self-reported and cover only standard text prompts, not image, video, or “deep research” style queries.

Do these numbers include the electricity used to train the AI model?
No. Every figure in this article covers only inference, meaning the electricity used each time someone asks a question. Training a large language model in the first place uses a separate and, by most independent estimates, much larger amount of electricity, spent once rather than per query.

Will AI’s electricity use keep growing?
The International Energy Agency projects global data center electricity consumption will more than double, from about 415 terawatt-hours in 2024 to around 945 terawatt-hours by 2030, with AI as the leading driver. This is a forecast, not a guaranteed outcome, and depends on how efficiently future AI models are built.

Sources

• Sam Altman, OpenAI, June 2025, via Data Center Dynamics
• Epoch AI, “How much energy does ChatGPT use?”, February 2025
• Google, Gemini energy footprint technical paper, August 2026
• International Energy Agency, “Energy and AI” and “Electricity 2026” reports

Figures in this article are current as of September 2026 and are self-reported by OpenAI and Google unless otherwise noted. Query volume, model efficiency, and hardware all change quickly in AI, so treat exact numbers as a snapshot rather than a permanent fact.