Artificial intelligence can answer questions, write articles, generate pictures, translate languages, and even create videos within seconds. But behind these seemingly simple activities are powerful computers working inside large facilities that require enormous amounts of electricity.
Unlike a person who can think without plugging their brain into a power socket, AI systems depend on electronic equipment that must be powered, cooled and maintained.
The growing use of AI has therefore created an important question: why does a technology that appears to exist mainly on the internet need so much electricity?
The short answer:
- AI needs electricity to power the powerful computer chips that process information and generate responses.
- Training AI models requires computers to perform enormous numbers of calculations over extended periods.
- Every time people use an AI service, computers must perform additional calculations to produce answers.
- Data centres also consume electricity for cooling, networking, storage and backup systems.
- The growing number of AI users and increasingly demanding features are increasing total electricity consumption.
- Although AI systems are becoming more energy-efficient, the rapid growth in their use can outweigh some of those efficiency improvements.
1. Where does AI get its electricity?
Most large AI systems operate inside facilities called data centres.
A data centre is a building or group of buildings containing powerful computers, storage equipment and networking systems. These facilities allow online services to operate around the clock.
When someone asks an AI chatbot a question, the request travels through the internet to computers in a data centre. Those computers process the request and send the response back to the user’s device.
The same infrastructure supports many other online services, including websites, email platforms, video streaming, cloud storage and online banking.
However, AI introduces additional electricity demands because some AI systems require specialised computer chips that perform enormous numbers of mathematical calculations.
The International Energy Agency (IEA) explains that servers account for approximately 60% of electricity consumption in modern data centres on average, although the proportion varies between facilities. Cooling, storage, networking and other equipment also require electricity. [1]
2. Why does artificial intelligence need powerful computer chips?
A normal computer can perform everyday activities such as browsing the internet, editing documents and watching videos.
However, developing and operating advanced AI models requires computers capable of performing huge numbers of calculations very quickly.
These calculations help AI systems recognise patterns, understand instructions, analyse information and generate responses.
One important piece of equipment is the graphics processing unit (GPU).
A GPU is a specialised computer chip originally developed to handle the calculations needed to create images and graphics. Its ability to perform many calculations at the same time also makes it useful for artificial intelligence.
Another important component is the central processing unit (CPU), which handles general instructions and coordinates many of the computer’s activities.
AI data centres often use large numbers of specialised processors working together.
Think of a school examination in which thousands of mathematical questions must be solved. One person working alone would take a long time, but hundreds of people working on different questions simultaneously could finish much faster.
GPUs work in a somewhat similar way. They can handle many calculations in parallel, meaning that several calculations are performed at the same time.
However, performing these calculations requires electrical power. The more powerful chips a facility operates, and the more intensively they are used, the more electricity it may consume. [1, 2]
3. Training an AI model consumes enormous amounts of energy
Before an advanced AI model can answer questions, it must first be developed through a process called training.
Training involves exposing an AI model to large amounts of information so that it can learn patterns and relationships.
For example, a language model may process large collections of written material to learn how words, sentences and ideas are connected.
During training, the system repeatedly makes calculations, compares its predictions with expected results and adjusts internal mathematical values to improve its performance.
These internal values are called parameters. They help determine how the model responds to different inputs.
This process can involve billions or even trillions of parameters, depending on the model.
Training can require thousands of specialised processors operating together for extended periods. These processors need electricity throughout the process, not just when someone is actively using the AI.
The computers also generate heat while working, creating an additional need for cooling.
Training is therefore one important source of AI’s electricity demand, particularly when developing large and advanced models. [1, 2]
4. AI also consumes electricity every time someone uses it
Training is not the only stage that requires energy.
After an AI model has been developed, it must continue performing calculations whenever people use it.
This process is called inference. In simple terms, inference is what happens when an already-trained AI model receives a question or instruction and produces an answer.
For example, when someone asks an AI chatbot to write a 500-word article, the system processes the instruction and generates the requested text.
When another person asks the same chatbot to translate a document, the system performs another set of calculations.
Every request requires computing resources and therefore consumes electricity.
The amount depends on several factors, including the model being used, the length of the response, the complexity of the task and the equipment running the system.
A simple question may require relatively little computing power. A complicated task involving extensive reasoning, large documents or several connected tools may require considerably more.
When millions of people use AI services throughout the day, these individual requests add up to substantial electricity consumption. [2]
5. Why generating an AI video can require more electricity than answering a simple question
Not all AI activities require the same amount of computing power.
Generating a short text response is generally less demanding than creating a detailed video containing moving objects, changing backgrounds, lighting and other visual elements.
This is because video generation requires the system to create and process many individual images, known as frames, and ensure that they work together as a moving sequence.
Other demanding activities include complex reasoning, generating high-quality images, analysing long documents and operating AI agents.
An AI agent is a system that can carry out several connected tasks with limited human direction. For example, it might search for information, compare results, organise findings and prepare a report.
Each additional step may require further processing.
The IEA’s April 2026 analysis explains that some newer AI applications, including video generation, advanced reasoning and agent-based tasks, can consume hundreds or even thousands of times more energy per request than simple text generation.
This is why it is misleading to assume that every AI question uses the same amount of electricity. [2]
6. Why do AI data centres need so much cooling?
Electricity does not simply disappear after powering a computer.
When computer chips process information, some of the electrical energy is converted into heat. If this heat is not removed, equipment can become too hot and may slow down, malfunction or suffer damage.
AI processors can generate particularly large amounts of heat because they perform intensive calculations.
Data centres therefore use cooling systems to maintain suitable operating temperatures.
Some facilities use powerful fans and air-conditioning equipment to move heat away from servers. Others use liquid cooling, where a circulating liquid carries heat away from equipment.
Liquid cooling is a method of removing heat using a liquid rather than relying entirely on moving air.
It is increasingly useful for advanced AI systems because some processors produce more heat than conventional air-cooling arrangements can efficiently manage.
Cooling systems also consume electricity. According to the IEA, cooling can account for around 7% of electricity consumption in efficient large-scale data centres, but more than 30% in some less-efficient facilities. [1]
This means the electricity used by an AI facility is not limited to the chips performing calculations. Keeping those chips at safe temperatures is another important part of the energy demand.
7. How much electricity do AI data centres actually consume?
The growing electricity demand of data centres has become an important subject for energy researchers and governments.
According to the IEA’s 2026 analysis, data centres around the world consumed approximately 485 terawatt-hours (TWh) of electricity in 2025.
A terawatt-hour is a unit used to measure large quantities of electricity. One terawatt-hour equals one billion kilowatt-hours.
To understand the scale, the electricity consumed by data centres worldwide in 2025 was equivalent to the annual electricity consumption of many millions of households.
The IEA projects that global data centre electricity consumption could reach approximately 950 TWh by 2030, roughly twice the 2025 level.
It also expects electricity consumption from AI-focused data centres to triple between 2025 and 2030. [2]
It is important to understand that these figures cover data centres, not AI alone. Data centres also support websites, cloud computing, streaming, business applications and other digital services.
AI is a major driver of the projected growth, but it is not responsible for every unit of electricity consumed by these facilities.
The growth is also uneven across regions. The United States, China and Europe currently account for major shares of data centre electricity consumption, while Southeast Asia is emerging as an important growth area. [1, 2]
8. Why is AI electricity consumption increasing so quickly?
There are several reasons.
More people are using AI
AI chatbots and other AI-powered tools are becoming part of everyday activities. People use them to write, study, research, translate, create images and solve problems.
As the number of users increases, data centres must handle more requests.
AI systems are becoming more advanced
Newer AI services can perform tasks beyond ordinary text generation. They can analyse images, create videos, process audio, reason through complex instructions and interact with other applications.
These features can require more computing power.
Businesses are adopting AI
Companies are introducing AI into customer support, software development, financial services, manufacturing, marketing and other operations.
Some businesses also develop their own AI systems, increasing demand for computing infrastructure.
AI models are becoming larger and more capable
Developing advanced models can require more powerful processors, larger computer systems and additional electricity.
However, bigger models are not automatically necessary for every task. Smaller models can sometimes perform routine activities with fewer computing resources.
AI services are available around the clock
Many major AI platforms operate continuously because users expect them to be available whenever needed.
Data centres must therefore maintain reliable infrastructure, even when individual users are not actively sending requests. [1, 2]
9. Does one AI question use as much electricity as charging a phone?
Claims about the electricity used by a single AI question can be misleading.
There is no single figure that applies to every AI request.
A simple text question sent to a relatively efficient model may require a small amount of electricity. A request to generate a long video or perform a complicated series of tasks may require considerably more.
The result also depends on the computer chips, software, data centre design and cooling arrangements.
The IEA’s 2026 report highlights an important development: electricity consumption per AI task has fallen rapidly as computer hardware and software have improved.
It says simple text queries now typically consume less electricity than running a television over the same period of time.
However, more demanding applications can use hundreds or thousands of times more energy per request than simple text generation. [2]
This means it is not accurate to claim that every AI question consumes the same amount of electricity as charging a phone, boiling water or running a household appliance.
Such comparisons need to specify the particular AI model, task, equipment and measurement method.
10. Is AI becoming more energy-efficient?
Yes. Improvements in computer hardware and software are helping AI systems perform tasks using less electricity.
One important development is the creation of more efficient processors.
Newer chips can perform more calculations using a given amount of electrical power. This is particularly important because advanced AI systems require large numbers of calculations.
Software improvements also help.
Developers can make models smaller, reduce unnecessary calculations and improve the way computing tasks are distributed across processors.
Another technique is model quantisation. This involves reducing the numerical precision used to represent information inside an AI model, which can reduce memory requirements and computing costs.
For example, a model may be able to perform a particular task using smaller numerical values without a major reduction in its usefulness.
Developers also use a method called model distillation, in which a smaller model is trained to reproduce much of the useful behaviour of a larger model.
This can make some AI applications cheaper and less demanding to operate.
The IEA reported in April 2026 that energy use per AI task had been falling by at least an order of magnitude annually in recent years. An order of magnitude means a factor of ten. [2]
However, greater efficiency does not automatically mean total electricity consumption will fall.
If AI becomes cheaper and easier to use, more people and organisations may use it. This can increase the total number of calculations being performed.
This is one reason why electricity demand can continue increasing even as individual AI tasks become more efficient.
11. Does AI electricity consumption affect ordinary people?
It can, particularly in places where large data centres are concentrated.
Data centres require reliable electricity supplies. When several large facilities operate in the same area, they can place significant demands on local electricity networks.
An electricity grid is the network of power stations, transmission lines, substations and distribution systems that delivers electricity to homes and businesses.
If a region does not have enough electricity generation or grid capacity, new data centres may require additional infrastructure.
This can involve building power plants, upgrading transmission lines, installing transformers or expanding substations.
A transformer is electrical equipment that changes voltage levels so electricity can be transmitted and distributed safely and efficiently.
These developments require investment, planning and time.
There can also be concerns about who pays for the additional infrastructure. If costs are not allocated carefully, households and other businesses could face some of the financial burden.
However, data centres do not automatically cause electricity prices to rise everywhere. The effect depends on local electricity supply, available infrastructure, investment decisions and government policies.
The IEA’s 2026 analysis explains that data centres can create affordability challenges because their electricity demand is large, concentrated and capable of growing quickly. It also notes that appropriate planning and grid investment can help manage these effects. [2]
12. Does AI’s electricity consumption contribute to climate change?
It can, depending partly on how the electricity is generated.
Electricity can come from different sources, including coal, natural gas, nuclear power, hydropower, wind and solar energy.
When a data centre uses electricity generated from fossil fuels, such as coal or natural gas, its electricity demand can contribute indirectly to greenhouse gas emissions.
Greenhouse gases are gases that trap heat in the Earth’s atmosphere. Carbon dioxide is one of the most important examples.
However, electricity generated from wind, solar, hydropower and nuclear energy generally has much lower operational greenhouse gas emissions than electricity generated from fossil fuels.
This means two data centres consuming the same amount of electricity may have different carbon footprints depending on their energy sources and other operating conditions.
A carbon footprint is the total greenhouse gas emissions associated with an activity, product or organisation.
Some technology companies purchase renewable electricity or sign agreements with renewable energy producers to support their operations.
A power purchase agreement is a contract through which an organisation agrees to buy electricity from an energy producer, often over an extended period.
However, purchasing renewable electricity does not automatically mean every data centre is powered entirely by renewable energy at every moment. The timing, location and availability of electricity matter.
The IEA’s 2026 analysis identifies growing electricity demand and associated emissions as important issues for the expansion of data centres. [2]
13. Can renewable energy power AI data centres?
Yes. Renewable energy can supply electricity to AI data centres.
Solar and wind power are among the options being used or considered by technology companies seeking to reduce their dependence on fossil fuels.
Hydropower can also provide electricity in regions with suitable water resources.
However, renewable energy sources have different operating characteristics.
Solar panels generate electricity when sunlight is available, while wind turbines depend on wind conditions. Electricity demand from data centres, on the other hand, can continue throughout the day and night.
This creates a need for reliable electricity supplies even when renewable generation is low.
One solution is battery storage, which stores electricity for use at a later time.
Another is combining renewable energy with other sources of electricity, including nuclear power, hydropower and, where used, natural gas.
Technology companies are also exploring advanced nuclear technologies and improved energy storage systems.
The IEA reported in April 2026 that technology companies accounted for around 40% of corporate power purchase agreements for renewable energy signed in 2025. [2]
The challenge is not simply finding clean energy. It is ensuring that enough reliable electricity can reach data centres when it is needed.
14. Could AI eventually help reduce electricity consumption?
Yes. Although AI itself requires electricity, it can also help other industries use energy more efficiently.
For example, electricity companies can use AI to analyse patterns in electricity demand and improve their forecasts.
Better forecasting can help operators prepare for periods when households and businesses are likely to consume more electricity.
AI can also help identify faults in electrical equipment, monitor industrial machines and improve the operation of heating and cooling systems.
In manufacturing, AI can analyse production processes and identify opportunities to reduce unnecessary energy use.
In buildings, AI-powered systems can adjust air conditioning and lighting based on occupancy, temperature and other conditions.
However, these benefits depend on how the technology is designed and used.
An AI system that saves electricity in one process may consume electricity through the computers and infrastructure needed to operate it.
The overall benefit therefore depends on whether the energy saved is greater than the energy required to run the AI system.
The IEA’s 2026 analysis identifies opportunities for AI to improve efficiency in energy-intensive industries, while also recognising that the technology’s own electricity requirements are increasing. [2]
15. What is being done to reduce the electricity AI consumes?
Researchers, technology companies and energy providers are working on several approaches.
Developing more efficient chips: New processors aim to perform more calculations while using less electricity.
Building smaller AI models: Smaller models can handle many everyday tasks without requiring the computing resources of the largest systems.
Improving cooling systems: Better cooling designs can reduce the electricity required to maintain safe operating temperatures.
Using computing resources more efficiently: AI developers can distribute tasks across available processors and avoid unnecessary calculations.
Locating data centres strategically: Building facilities where reliable electricity and suitable infrastructure are available can reduce pressure on already-constrained areas.
Using renewable energy and storage: Solar, wind, hydropower and battery systems can help supply electricity while reducing dependence on fossil fuels.
Making data centres more flexible: Some computing tasks can be moved to different times or locations when electricity is more available.
These approaches are important because the future electricity demand of AI will depend not only on how many people use it, but also on how efficiently the technology is developed and operated. [1, 2]
16. What could happen if AI electricity demand continues growing?
The expansion of AI could influence how countries plan their electricity systems.
Governments and energy companies may need to build additional power generation capacity, strengthen electricity grids and improve energy storage.
The demand for specialised electrical equipment, including transformers, cables and power-control systems, may also increase.
The IEA’s April 2026 analysis estimated that global data centre electricity consumption could rise from approximately 485 TWh in 2025 to 950 TWh in 2030. It also projected that electricity use from AI-focused data centres could triple over that period. [2]
These are projections, not guaranteed outcomes. Actual consumption will depend on how quickly AI adoption grows, how efficiently systems operate, the availability of equipment and the ability of electricity networks to support new facilities.
There may also be differences between countries.
Regions with reliable electricity supplies, suitable infrastructure and favourable investment conditions may attract more data centre development.
Other regions may face difficulties if electricity generation and transmission infrastructure cannot keep pace with demand.
For countries across Africa and Asia, the issue is not simply whether AI data centres can be built. It is also whether the necessary electricity infrastructure can be developed without undermining access, affordability and reliability for existing users.
17. Myths vs facts about AI and electricity
| Claim | Status | What is actually true |
|---|---|---|
| Every question asked of AI consumes a huge amount of electricity. | Myth | Electricity use varies considerably. Simple text requests generally require less energy than complex tasks such as video generation. |
| AI only uses electricity when someone is chatting with it. | Myth | Data centres also consume electricity for training, storage, networking, cooling and maintaining infrastructure. |
| Training an AI model is the only stage that consumes electricity. | Myth | AI systems also consume electricity whenever they process requests and generate responses. |
| AI computers produce heat while operating. | True | Electrical energy used by computer equipment is ultimately converted into heat that must be removed. |
| All AI models consume the same amount of electricity. | Myth | Energy use depends on model size, hardware, software, task complexity and operating conditions. |
| AI is becoming more energy-efficient. | True | Improvements in hardware and software are reducing electricity consumption per task. |
| Greater efficiency guarantees that total AI electricity consumption will fall. | Myth | Increased usage and more demanding applications can outweigh efficiency improvements. |
| AI data centres can operate using renewable energy. | True | Renewable sources can supply data centres, although reliable electricity is still needed when generation fluctuates. |
| AI electricity demand can affect local electricity networks. | True | Large data centres can create concentrated demand that requires additional grid capacity and investment. |
| AI will eventually consume all the world’s electricity. | Myth | Current projections show substantial growth, but not anything close to consuming all global electricity. |
Bottom line
AI needs so much electricity because it depends on powerful computers that perform enormous numbers of calculations. Those computers require electricity to train models, answer questions, generate images and videos, and operate the digital infrastructure behind AI services. Additional electricity is needed to cool equipment, store information and keep data centres running reliably.
The important development is that AI is becoming more efficient, but it is also being used by more people and for increasingly demanding tasks. As a result, the total electricity required by the industry is growing.
The challenge is to develop AI in a way that makes its benefits worthwhile while ensuring that the electricity needed to power it is reliable, affordable and increasingly clean.
Disclaimer: This article is for general education only and does not replace professional advice relevant to the topic. If you notice an error or would like to send feedback or corrections, please contact us.


