Verify any claim · lenz.io
Claim analyzed
Tech“Twenty ordinary text prompts to an artificial intelligence system consume approximately 6 watt-hours of data-centre electricity, assuming electricity use of 0.3 watt-hours per prompt.”
Submitted by Bright Sparrow 1b63
The conclusion
Open in workbench →The calculation is sound under its stated assumption: 20 prompts multiplied by 0.3 watt-hours equals 6 watt-hours. Recent estimates around 0.24–0.34 watt-hours for typical text prompts make 0.3 watt-hours a reasonable approximation, although actual consumption varies substantially by model, response length, hardware, and accounting boundary.
Caveats
- The 0.3-watt-hour figure is an approximate typical value, not a universal rate for every AI prompt.
- Energy use varies with model, response length, decoding strategy, hardware efficiency, and workload.
- Estimates may use different boundaries, such as GPU-only consumption versus full data-centre overhead.
Get notified if new evidence updates this analysis
Create a free account to track this claim.
Sources
Sources used in the analysis
Their findings show that earlier research presented by De Vries et al. 17 estimating ChatGPT’s energy usage at about 3 watt-hours per query was significantly exaggerated, with the more accurate estimation being closer to 0.3 watt-hours per query.
Each interaction consumes energy—about 0.34 watt-hours per prompt.
This is consistent with recent estimates for a typical chatbot query in a highly-optimized deployment: 0.3 Wh for a typical GPT-4o query of 500 output tokens (heuristic estimate based on FLOPs of a 200B model) [10], 0.421 Wh for GPT-4o query with 300 output tokens (based on user-side token throughput) [8], 0.34 Wh for the “average energy use of a ChatGPT query”, as disclosed by Sam Altman [11], and 0.24 Wh for “the typical Gemini Apps query” [2].
Their findings show that earlier research presented by [17] estimating ChatGPT’s energy usage at about 3 watt-hours per query was significantly exaggerated, with the more accurate estimation being closer to 0.3 watt-hours per query.
The calculation assumes that the prompt was processed by the GPT-4o model, which is believed to use a mixture-of-experts architecture with about 100 billion active parameters, running on NVIDIA H100 GPUs. The energy consumption of a typical ChatGPT prompt is estimated to be approximately 0.3 Wh.
In 2024, a single query on an advanced generative AI model like ChatGPT required an estimated 2.9 watt-hours of electricity, nearly 10 times the 0.3 watt-hours needed for a conventional Google search. … Newer measurements suggest median energy per text query has fallen to 0.24-0.3 watt-hours, although this can be much higher for long reasoning or multimodal prompts.
Using OpenAI’s latest default model for ChatGPT, GPT-4o, as a reference, Epoch found the average ChatGPT query consumes around 0.3 watt-hours — less than many household appliances.
The analysis, focused on serving AI at large scale, finds that a typical AI query to some of the largest and most capable LLMs uses between 0.16 and 0.60 watt-hours of electricity, depending on the length of the query, the LLM used, and datacenter specifications.
In total, the median prompt—one that falls in the middle of the range of energy demand—consumes 0.24 watt-hours of electricity, the equivalent of running a standard microwave for about one second.
In total, the median prompt—one that falls in the middle of the range of energy demand—consumes 0.24 watt-hours of electricity, the equivalent of running a standard microwave for about one second.
Estimates of ChatGPT’s per-query energy consumption vary wildly. We used the figure of 0.34 watt-hours that OpenAI’s Sam Altman stated in a blog post without supporting evidence.
A typical Gemini text query uses 0.24 watt-hours (Wh) of energy, emits 0.03 grams of carbon dioxide equivalent (gCO2e), and consumes 0.26 milliliters — or about five drops — of water. … By comparison, the average ChatGPT query uses 0.34 Wh and about one fifteenth of a teaspoon of water, Sam Altman, CEO of ChatGPT-maker OpenAI, has written.
We find that typical ChatGPT queries using GPT-4o likely consume roughly 0.3 watt-hours, which is ten times less than the older estimate.
Using this methodology, we estimate the median Gemini Apps text prompt uses 0.24 watt-hours (Wh) of energy, emits 0.03 grams of carbon dioxide equivalent (gCO2e), and consumes 0.26 milliliters (or about five drops) of water1 — figures that are substantially lower than many public estimates.
The calculation assumes that the prompt was processed by the GPT-4o model, which is believed to use a mixture-of-experts architecture with about 100 bil lion active parameters, running on NVIDIA H100 GPUs. The energy consumption of a typical ChatGPT prompt is estimated to be approximately 0.3 Wh.
In 2025, [it released](https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference/) energy estimates for its Gemini model. It estimated the _median_ text-based query (basically asking Gemini a text question) consumed around 0.24 watt-hours (Wh) of electricity. … The CEO of OpenAI, Sam Altman, also [previously wrote](https://blog.samaltman.com/the-gentle-singularity) that an “average query” on ChatGPT consumed around 0.34 Wh (but without a detailed breakdown of where this number comes from). Epoch AI also provided its own independent [estimate of](https://epoch.ai/gradient-updates/how-much-energy-does-chatgpt-use) around 0.3 Wh for a “typical” ChatGPT query.
"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," Altman said in a blog post.
We find that typical ChatGPT queries using GPT-4o likely consume roughly 0.3 watt-hours, which is ten times less than the older estimate.
We find that typical ChatGPT queries using GPT-4o likely consume roughly 0.3 watt-hours, which is ten times less than the older estimate.
Epoch AI estimates a typical ChatGPT query running on GPT-4o uses roughly 0.3 watt-hours, about ten times less than an earlier and widely cited 3-watt-hour figure, a gap they attribute to more efficient models and more realistic assumptions about how long a typical answer runs.
Even a 0.42 Wh short query, when scaled to 700M queries/day, aggregates to annual electricity comparable to 35,000 U.S. homes, evaporative freshwater equal to the annual drinking needs of 1.2M people, and carbon emissions requiring a Chicago-sized forest to offset.
Generating large amounts of text in experimental conditions can consume as much electricity as an LED bulb.
While a single short GPT-4o query consumes 0.42 Wh, scaling this to 700 million queries/day results in substantial annual environmental impacts.
For instance, a ChatGPT query is estimated to consume about 2.9 Wh, nearly ten times the 0.3 Wh of a regular Google search [14].
Knowing the batch size, we can derive the energy cost of a single prompt. We assume this particular scenario, as the number of operations is the same for each of the prompts, i.e., all prompts go through the LLM layers the same amount of time, generates the same amounts of tokens.
Our monitoring tools measure the power consumption P of a process in microwatts (µW). We then compute the energy consumption E by integrating the power consumption over the duration t, applying the trapezoidal rule. With t expressed in seconds, we convert E
Results show per text prompt energy consumption of 0.24 Wh. … Energy: 0.24 Wh per prompt
Per prompt, that works out to roughly 150 Wh, about 600 times as much as a median chat prompt.
Studies on inference energy are much rarer because a single inference operation consumes considerably less energy and resources.
The rapid expansion of artificial intelligence (AI), especially Large Language Models (LLMs) such as GPT-3 and Gemini on which the now well-known ChatGPT AI and Gemina assistant systems are based, has raised concerns about whether and how this new technology may impinge on the ability of the United States to meet its zero-carbon electricity goals.
A single AI text prompt uses roughly 0.2–0.3 watt-hours of electricity — far less than the alarming numbers that circulated in 2023, but the fleet-level total is what’s exploding.
An independent estimate from Epoch AI lands in the middle at roughly 0.3 Wh. … Run the yearly math: 20 prompts a day, every day, is about 2.5 kWh per year.
As these services scale, a fundamental systems ques tion remains largely underexplored: “What is the power consumption of serving an LLM prompt?”
System-level and datacenter-level contributions, including CPU execution, host memory, networking, storage, power-supply losses, cooling, and power usage effectiveness (PUE), are outside the scope of this work and are not included in the reported estimates.
How Much Energy Does Your AI Prompt Use?
Big model has emerged as a new research paradigm that can be applied to various down-stream tasks with only minor effort for domain adaption.
Article How Much Energy Does One AI Prompt Really Use? The Real 2026 Numbers Verified numbers on AI energy consumption per prompt: 0.34 Wh for ChatGPT, 0.24 Wh for Gemini, the physics floor, and why totals still explode.
What do you think of the claim?
Your challenge will appear immediately.
Challenge submitted!
For developers
This same pipeline is available via API.
Verify your AI's output programmatically.
/extract pulls claims from text ·
/verify returns sourced verdicts ·
/ask answers follow-up questions.
Continue your research
Verify a related claim next.
Debate
Two AI advocates debated this claim using the research gathered.
Argument for
Multiple high-quality sources including Source 1 (nature.com), Source 3 (arxiv.org), Source 5 (arxiv.org), Source 7 (techcrunch.com), and Source 13 (epoch.ai) establish that ordinary AI text prompts consume approximately 0.3 watt-hours of data-centre electricity each. Multiplying this consensus per-prompt figure by twenty therefore yields exactly 6 watt-hours, confirming the claim as arithmetically and empirically true.
The Proponent commits a fallacy of equivocation by conflating baseline GPU inference energy with total data-centre electricity, ignoring that the 0.3 watt-hour figure excludes massive facility overheads. As explicitly noted in Source 35 (arxiv.org), these baseline estimates omit system-level contributions like cooling and power usage effectiveness (PUE), meaning the actual data-centre electricity required for twenty prompts is substantially higher than 6 watt-hours.
Argument against
While limited evidence suggests a baseline GPU energy draw of 0.3 watt-hours per prompt, the claim falsely equates this partial metric with total data-centre electricity consumption. As Source 35 explicitly clarifies, these baseline estimates exclude massive system-level and datacenter-level contributions like cooling, networking, and power usage effectiveness (PUE), meaning the actual data-centre electricity required for twenty prompts is substantially higher than 6 watt-hours.
The Opponent cherry-picks Source 35's narrow methodological caveat while ignoring the broader consensus in Sources 1, 3, 5, 7, 8, and 13 that already frame approximately 0.3 watt-hours as the full data-centre electricity cost of a typical prompt, inclusive of optimized deployment realities. This selective reading commits a fallacy of incomplete evidence, as the claim's arithmetic (20 × 0.3 = 6) rests on those validated total-usage figures rather than any partial GPU-only metric.
Panel Review
3 specialized AI experts evaluated the evidence and arguments.
Reviewer 1 — The Logic Examiner
The evidence consistently supports the premise that a typical AI text prompt consumes approximately 0.3 watt-hours of electricity, as cited by multiple sources including nature.com, arxiv.org, and techcrunch.com. The arithmetic logically follows that twenty such prompts would consume approximately 6 watt-hours, making the claim's conclusion sound based on the provided data.
Reviewer 2 — The Source Auditor
Multiple high-quality, independent sources — Source 1 (nature.com), Source 3 (arxiv.org), Source 5/15 (Google's own methodology paper), Source 7 (TechCrunch reporting Epoch AI's analysis), Source 13/18/19 (Epoch AI), and Source 8 (Microsoft) — converge on approximately 0.3 Wh per typical text prompt as an estimate of the full data-centre electricity cost (not merely GPU compute), with Google's and Microsoft's figures explicitly described as end-to-end or datacenter-inclusive estimates (0.24-0.6 Wh range). The Opponent's reliance on Source 35 is weak since that source explicitly states system-level/datacenter contributions are 'outside the scope' of its own narrower analysis, which does not establish that the widely-cited 0.3 Wh figures from Google, Microsoft, and Epoch AI exclude such overhead — in fact those sources (5, 8, 14, 15) claim to already include full deployment-level costs; thus 20 x 0.3 Wh = 6 Wh is well-supported as a reasonable approximation, though the true figure has meaningful uncertainty (0.16–0.6 Wh per prompt) and reasonable sources like Brookings and Google put median slightly lower (0.24 Wh), making 6 Wh a defensible but not perfectly precise midpoint estimate.
Reviewer 3 — The Precision Analyst
The claim's arithmetic (20 × 0.3 Wh = 6 Wh) is exact, and multiple sources (1, 3, 5, 7, 13, 16) directly support the assumed ~0.3 Wh per ordinary text prompt as a data-centre electricity figure for typical chatbot queries, with nearby estimates of 0.24–0.34 Wh confirming the approximation. As worded with its explicit assumption, the claim is therefore true at the stated strength.