The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Every AI feature uses electricity when it runs, but there is no single energy cost for “one AI query.” A short text reply, a long reasoning task, and a video-generation request can use very different amounts. The estimate also changes depending on whether it counts only the accelerator or the full serving system, including host computers, idle capacity, and data-center overhead. Per-request efficiency is only half the picture: total electricity use can still rise as more people use AI and adopt energy-intensive features.
How much electricity does one AI query use?
Published estimates are useful examples, not a universal tariff. Google reported that the median text-generation prompt in Gemini Apps used 0.24 watt-hours (Wh) in May 2025, based on the company’s comprehensive production measurement boundary. Microsoft Research modeled a median of 0.34 Wh per query for frontier-scale models above 200 billion parameters under stated H100-node workload assumptions. These are not directly comparable provider scores: one is a company’s production disclosure for a named product and prompt type, while the other is a model-based estimate for a different workload and system.
For scale, 1 Wh is the energy used by a 1-watt device running for one hour. The figures above describe electricity per request under their respective methods; they do not tell you the total electricity used to operate an AI service over a day or year.
Google’s May 2025 disclosure is a point-in-time, provider-reported estimate for the median Gemini Apps text-generation prompt. The company says it does not represent all prompts, is not indicative of future performance, and has not been independently verified. Its associated paper reports 0.24 Wh using a comprehensive boundary, versus 0.10 Wh under a narrower method for the same product. That difference shows why a number without its measurement boundary can mislead. Google’s explanation of the Gemini Apps estimate and its technical paper describe the methods.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallMicrosoft Research’s 0.34 Wh median estimate has an interquartile range of 0.18–0.67 Wh. It models frontier-scale models running on an H100 node, with workload, GPU-utilization, and data-center power-usage-effectiveness assumptions. It is not a measurement of a named consumer feature. Microsoft Research’s analysis explains the scenario and its assumptions.
Why do some AI features use more energy than others?
The job matters. The International Energy Agency (IEA) says video generation, reasoning, and agentic tasks can consume hundreds or thousands of times more energy per query than simple text generation. This is a broad assessment, not a universal lookup table for individual features. A feature’s workload can also vary from one request to another: a long input or output, for example, can require more processing than a short exchange.
In a modeled test-time-scaling scenario, Microsoft Research estimates a median of 4.32 Wh when a frontier-scale model uses 15 times more tokens than in its baseline case, where the median is 0.34 Wh. This illustrates how additional computation can change energy use substantially; it is not a universal measurement of all reasoning features.
Serving conditions matter as well. Hardware, batching, utilization, and provisioned capacity affect how electricity is attributed to a request. A calculation that includes only active accelerators will differ from one that also counts host CPU and memory, machines kept available but idle, and data-center overhead.
What a full-stack estimate can include
In Google’s May 2025 analysis of the median Gemini Apps prompt, the company attributes its 0.24 Wh full-stack total to active accelerator power (0.14 Wh, or 58%), host CPU and DRAM (0.06 Wh, or 25%), provisioned idle machines (0.02 Wh, or 10%), and data-center overhead (0.02 Wh, or 8%). These are the company’s reported components for that product and analysis, not a standard breakdown for other systems.
How to compare two energy estimates
Before treating two per-query numbers as comparable, check what each one actually describes:
- Task: Is it short text generation, a long-context response, reasoning, an agentic workflow, image creation, or video generation?
- System boundary: Does the figure include just the accelerator, or also CPU, memory, idle capacity, and data-center overhead?
- Workload and scale: What hardware, tokens per request, batching, and utilization are assumed? Is the number measured in production or modeled?
- Evidence and date: Is it a provider disclosure, a research model, or an observed system-wide statistic? When does it apply?
- For emissions estimates: What electricity mix, location, and time period were used to convert electricity into carbon emissions?
These distinctions explain why the two example estimates above should not be read as a head-to-head comparison of providers. The underlying tasks, boundaries, and evidence types differ.
What do energy, emissions, and water figures mean?
Energy use is not the same as a prompt’s carbon footprint. Google calculated 0.03 grams of carbon-dioxide equivalent (gCO2e) for its median Gemini Apps text prompt using the company’s 2024 average fleet-wide grid carbon intensity. It estimated 0.26 millilitres of water using its 2024 average fleet-wide water-usage effectiveness. These are calculations using fleet averages, not direct measurements of the local electricity or water impact of each individual prompt. Google also reported that the median prompt’s energy use fell 33-fold and its carbon footprint 44-fold between May 2024 and May 2025. That comparison applies to Gemini Apps text prompts, not all providers or AI workloads.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →At the industry level, the IEA estimated that data centers accounted for about 415 terawatt-hours (TWh), or 1.5% of global electricity consumption, in 2024. That total includes AI and non-AI workloads. The agency estimated around 180 million tonnes of indirect CO2 emissions from data-center electricity use that year; this also covers all workloads and excludes emissions from backup power generation. It is not an AI-only emissions figure. See the IEA’s 2025 Energy and AI report and its analysis of AI-related energy demand.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can improving efficiency reduce total AI electricity use?
It can reduce the electricity needed for an individual task, but that does not guarantee a drop in total demand. The IEA says energy use per AI task has fallen by at least an order of magnitude annually in recent years, driven by software and hardware advances. At the same time, it reports that global data-center electricity demand grew 17% in 2025, while demand from AI-focused data centers grew 50%.
The IEA’s 2026 report puts total data-center electricity consumption at 485 TWh in 2025 and projects 950 TWh in 2030, around 3% of global electricity demand. The 2030 figure is a projection, and the total covers data centers, not AI alone. In its 2025 report, the IEA had projected around 945 TWh for 2030; that earlier forecast was also a projection, not an observed result. These dated estimates should be understood in the context of their respective reports, not treated as measurements of AI’s share.
As the IEA puts it: “Measured per individual task, the energy efficiency of AI is improving at a rate unprecedented in energy history.” It also warns that energy-intensive applications are increasingly being launched and used, “such as those for video generation, reasoning and agentic tasks.” The practical implication is that both sides matter: efficiency per request and the number and type of requests being served. The IEA’s 2026 Key Questions on Energy and AI discusses these trends.
Free tools Windows power users keep installed
One-click scans. No signup required.
Can you measure the energy used by one AI feature at home?
Usually not with a household energy meter or smart plug. When a feature runs on a provider’s cloud servers, those devices measure your local equipment—not the portion of remote data-center electricity attributable to your request. Running an AI model locally is a different case: a meter can capture the electricity used by your own computer, but that measures local operation rather than a cloud feature.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
