When people use an AI chatbot, generate an image or ask an artificial intelligence system to analyse a document, the work happens inside powerful computers, many of which are located in large buildings called data centres.
Some of these facilities are so large that they resemble industrial factories rather than ordinary office buildings. They may occupy extensive land, contain thousands of computer servers and require their own electricity and cooling infrastructure.
But why does artificial intelligence need so much physical space? Could all those computers not fit inside a smaller building?
The answer involves more than the computers themselves. AI data centres must accommodate powerful processors, electricity distribution systems, cooling equipment, storage devices, networking systems, backup power and other infrastructure needed to keep everything operating reliably. [1, 2]
The short answer:
- AI data centres need space for thousands of powerful computers working together.
- Their computer equipment generates considerable heat, requiring extensive cooling systems.
- They need large electrical systems to deliver enough power safely and reliably.
- Backup generators and batteries help keep services running during electricity interruptions.
- High-speed networking equipment connects thousands of computers so they can work together.
- Operators often build large campuses because they expect demand to grow and need room for expansion.
- Not every data centre is enormous. Their size depends on their purpose, computing capacity and design.
1. What exactly is an AI data centre?
An AI data centre is a facility designed to house the computers and supporting equipment used to develop, train and operate artificial intelligence systems.
It is essentially a large computing centre.
Inside, there are rows of computer equipment arranged in metal structures called server racks. A server rack is a frame or cabinet that holds multiple computers and related equipment in an organised space.
A server is a computer designed to provide computing services, process information or store data for other computers and users.
Unlike an ordinary desktop computer, a server is generally designed to operate continuously and handle requests from many users or other systems.
AI data centres contain specialised servers equipped with powerful processors that can perform large numbers of calculations. They also contain storage systems, networking equipment, cooling systems and electrical infrastructure. [1]
Some modern facilities are specifically designed for AI and are sometimes described as AI factories. This term refers to data centres built to process enormous amounts of information and produce AI outputs at scale.
The word factory is used because these facilities take information as input, perform large amounts of computing work and produce results such as text, images, videos and predictions.
2. AI needs thousands of computers working together
One major reason AI data centres are large is the amount of computing equipment they must accommodate.
Training a powerful AI model can involve enormous numbers of mathematical calculations. A single computer may not be able to complete the work quickly enough.
Instead, developers connect many computers so that they can perform different parts of the task simultaneously.
This is called parallel computing, which means dividing a large computing task among several processors or computers so that they can work at the same time.
Think of a construction project that requires thousands of bricks to be laid. One worker could complete the job, but it would take much longer than using a team of workers who handle different sections simultaneously.
AI data centres use a similar principle. Thousands of processors can work together to train models or handle demanding computing tasks.
The computers must also communicate with one another quickly. This requires additional networking equipment and connections.
The more computing capacity an operator wants to provide, the more servers, racks, cables, networking equipment and supporting infrastructure may be required. [1, 2]
3. Why AI processors require so much space
Not all computers are equally powerful.
Advanced AI systems often use specialised chips called graphics processing units (GPUs). These chips can perform many mathematical calculations simultaneously, making them useful for training and operating AI models.
Modern AI processors are becoming increasingly powerful, but they also require more electricity and produce more heat.
To use them effectively, data centre operators often group several processors into specialised servers and install those servers together in racks.
The equipment inside these racks can be much more demanding than the hardware found in conventional data centres.
According to the International Energy Agency’s 2026 report, the power density of AI servers increased 11 times between 2020 and 2025. Power density means the amount of electrical power required within a given amount of physical space.
The IEA also explains that an advanced AI server rack, which is approximately the size of a large refrigerator, could by 2027 have a peak electricity demand equivalent to that of around 65 households. [1]
This illustrates an important point: a relatively small area can contain equipment requiring an enormous amount of electricity.
However, operators still need enough space around the equipment for cooling pipes, electrical connections, maintenance access and safe operation.
4. AI data centres need enormous cooling systems
One of the biggest reasons AI data centres require extensive infrastructure is heat.
Computer processors generate heat when they perform calculations. The more intensively they operate, the more heat they can produce.
If this heat is not removed, equipment can become too hot and may slow down, malfunction or suffer damage.
Data centres therefore require systems that continuously manage temperature and airflow.
Traditional facilities may use powerful fans, air-conditioning systems and specialised cooling equipment.
However, advanced AI processors can produce so much heat that ordinary air cooling may not be sufficient for some installations.
This is why many modern AI data centres are adopting liquid cooling.
Liquid cooling is a method of removing heat using a circulating liquid rather than relying entirely on air.
One approach is called direct-to-chip cooling. In this arrangement, cooling plates are placed close to hot computer processors, and liquid circulates through them to carry heat away.
The heat is then transferred to other cooling equipment before the liquid is circulated again.
This requires pipes, pumps, heat exchangers and other supporting equipment.
A heat exchanger is a device that transfers heat from one fluid or system to another without necessarily mixing them.
These cooling systems occupy physical space and must be designed alongside the computer equipment they support. [2, 3]
5. The building needs a powerful electricity system
An AI data centre cannot operate with the same electrical arrangement as a normal office.
Thousands of powerful processors, storage devices, cooling systems and networking components may need electricity simultaneously.
The facility therefore requires substantial electrical infrastructure to receive, distribute and control power.
This can include:
- Transformers that change electricity voltage to suitable levels.
- Switchgear that controls and protects electrical circuits.
- Distribution panels that send electricity to different parts of the facility.
- Power cables and busways that carry electricity to server racks.
- Backup batteries and generators that support operations during interruptions.
A transformer is electrical equipment that increases or reduces voltage so electricity can be transmitted and distributed safely.
Switchgear is equipment used to control, protect and isolate electrical circuits.
These systems are not simply accessories. They are essential to making sure that thousands of computers receive the power they need without exposing equipment to dangerous electrical faults.
The IEA explains that servers account for around 60% of electricity consumption in modern data centres on average, although this varies considerably. Cooling, storage, networking and other infrastructure account for the rest. [2]
The electrical systems needed to support such facilities can therefore occupy substantial space, sometimes in separate rooms or dedicated buildings.
6. Why do data centres have backup generators and batteries?
Many online services are expected to remain available even when the local electricity supply experiences an interruption.
Imagine a bank’s digital services becoming unavailable every time there is a power cut. Customers could lose access to important services, and business operations could be disrupted.
AI services can face similar problems.
Data centres therefore use backup power systems to help maintain operations when the main electricity supply fails.
One important component is an uninterruptible power supply (UPS).
A UPS is a system that provides immediate backup electricity, usually from batteries, when the main power supply is interrupted.
It helps prevent computers from shutting down suddenly while backup generators start or another power source becomes available.
Generators can provide electricity for longer periods, depending on their capacity and available fuel.
These systems require physical space for batteries, electrical equipment, generators, fuel storage and safety arrangements.
The IEA identifies UPS batteries and backup generators as important parts of data centre infrastructure because these facilities must meet demanding reliability requirements. [2]
7. Thousands of computers need to communicate quickly
AI data centres are not simply collections of independent computers.
Many AI tasks require different processors to exchange information continuously.
For example, when a large AI model is being trained, different processors may handle different parts of the calculations. They need to exchange results and coordinate their work.
This requires high-speed networking equipment.
Networking equipment includes devices such as switches and routers that connect computers and direct information between them.
A network switch connects devices within a network and helps information reach the correct destination.
A router directs information between different networks.
Large AI facilities may use specialised high-speed connections to reduce the time computers spend waiting for information from one another.
This delay is called latency. It refers to the time taken for information to travel from one point to another.
Low latency is important because delays in communication can slow down large computing tasks.
Data centres therefore need extensive cabling, network switches, optical connections and other communication equipment.
Optical connections use light to transmit information through fibre-optic cables, allowing large amounts of data to travel quickly over distances.
All this equipment requires space, power, cooling and maintenance access. [1, 2]
8. Why do some AI data centres occupy entire campuses?
Some operators do not build just one large building. They develop several data centre buildings across a shared site called a data centre campus.
A campus may contain separate buildings for computing equipment, electrical systems, cooling equipment, security and other operations.
There are several reasons for this arrangement.
Room for expansion
AI demand can grow quickly. An operator may build one facility and later add more buildings as computing requirements increase.
Different functions
Separating some electrical, cooling and computing systems can make the overall design easier to manage.
Reliability
A campus can be designed with alternative power routes, backup systems and separate equipment areas to reduce the risk that one failure affects every operation.
Construction planning
Building several facilities in stages can allow operators to add capacity without completing the entire campus at once.
Security
Large campuses can use controlled entry points, security barriers, monitoring systems and restricted areas to protect equipment and information.
The IEA’s 2026 analysis describes the rapid expansion of AI-focused facilities, noting that tracked AI factories more than tripled in capacity over the previous 18 months. [1]
This helps explain why some companies are investing in large sites rather than relying only on small computer rooms.
9. Why can’t all the computers simply be placed in one smaller building?
In principle, some data centres can be made smaller by using more compact equipment and more efficient designs.
However, reducing the building’s physical size is not always the best solution.
Packing more equipment into a smaller area can increase the amount of heat and electricity concentrated in that space.
This creates additional demands for cooling, power distribution, fire protection and maintenance access.
The building must also support the weight of heavy computer racks, batteries and other equipment.
In some facilities, liquid-cooling systems require extensive pipe networks and additional space for pumps and heat exchangers.
There are also practical requirements for staff to inspect, repair and replace equipment safely.
The challenge is therefore not simply fitting as many computers as possible into the smallest building.
Operators must balance computing capacity, reliability, cooling, safety, maintenance and future expansion. [2, 3]
10. Are all data centres as large as AI data centres?
No.
Data centres vary greatly in size and purpose.
A small organisation may operate a server room containing a limited number of computers. A university, hospital or business may operate its own data centre to support internal services.
Other organisations use colocation data centres, which are facilities where different businesses rent space, power and cooling for their own computer equipment.
At the larger end are hyperscale data centres.
A hyperscale data centre is a very large computing facility designed to support enormous numbers of users and large online services. Major cloud providers operate facilities of this kind.
Some hyperscale facilities also support AI workloads, while others focus on cloud computing, storage, websites and other digital services.
AI-focused facilities are becoming particularly demanding because they use large groups of powerful processors that require substantial electricity and cooling.
However, it would be inaccurate to assume that every data centre is an enormous AI factory. Their size depends on the services they provide and the amount of computing capacity they need. [1, 2]
11. Why do AI data centres need so much land?
The computers themselves are only part of the physical development.
A large data centre campus may need land for:
- Main buildings containing servers and storage equipment.
- Electrical substations and power distribution infrastructure.
- Cooling equipment and heat rejection systems.
- Backup generators and battery installations.
- Roads, loading areas and maintenance access.
- Security barriers and controlled entry points.
- Additional buildings planned for future expansion.
An electrical substation is a facility that helps control electricity voltage and distribute power through the electricity network.
Some data centres also require water infrastructure, depending on the cooling method used.
For example, certain cooling systems use water to remove heat, while others rely more heavily on air or closed-loop liquid systems.
A closed-loop system circulates the same cooling liquid through equipment rather than continuously using fresh water for the cooling process. However, the facility may still require water for other purposes, depending on its design.
The amount of land required therefore depends on the facility’s capacity, cooling design, electrical supply arrangements and expansion plans. [2, 3]
12. Why are companies spending so much money building AI data centres?
AI has become an important part of technology companies’ plans for developing new products and services.
Companies need computing infrastructure to train their own models, operate AI assistants, provide cloud services and support businesses that use AI applications.
This has created demand for specialised chips, servers, buildings, electricity connections, cooling equipment and networking systems.
The International Energy Agency reported in its April 2026 analysis that capital expenditure by five major technology companies exceeded $400 billion in 2025 and was expected to increase by a further 75% in 2026.
Capital expenditure means money spent on long-term assets such as buildings, equipment and infrastructure. [1]
The same report explains that data centre electricity consumption grew by 17% globally in 2025, while electricity consumption from AI-focused data centres increased by 50%.
These figures illustrate the rapid expansion of the infrastructure supporting AI. [1]
However, not every proposed project will necessarily be completed. Construction depends on financing, access to electricity, equipment availability, planning approval and expected demand.
13. Does building a larger data centre automatically make AI more powerful?
No.
The physical size of a data centre does not directly determine how intelligent an AI model is.
AI performance depends on several factors, including the model’s design, training data, computing methods, software and hardware.
A larger facility can provide more computing capacity, allowing developers to train larger models or serve more users.
However, simply adding more buildings or computers does not automatically produce a better AI system.
The equipment must be properly connected, the software must use it effectively and the computing resources must be suitable for the task.
A smaller, well-designed AI model can sometimes perform a particular job more efficiently than a much larger model.
The size of the building is therefore better understood as a measure of the infrastructure available, not a direct measure of intelligence.
14. What problems can large AI data centres create for nearby communities?
Large data centres can bring investment, construction work, technical jobs and demand for local services.
However, they can also create challenges for the areas where they are built.
Pressure on electricity supply
A large facility may require a substantial electricity connection. If several facilities are built in the same area, they can increase demand on the local electricity network.
Water demand
Some cooling systems use water to remove heat. The effect on local water resources depends on the cooling design, climate, water source and facility operations.
Noise
Generators, fans, pumps and other equipment can produce noise. Proper design and sound-control measures are important, particularly near residential areas.
Land use
Large campuses require land that could otherwise be used for housing, agriculture, industry or other development.
Construction and infrastructure
Data centre projects may require new roads, electrical substations, transmission lines and other infrastructure.
These effects are not identical everywhere. They depend on the facility’s design, location, energy supply and relationship with the surrounding community.
The IEA identifies electricity access, infrastructure bottlenecks, affordability and community acceptance as important issues in the expansion of data centres. [1]
15. Can future AI data centres become smaller?
Some parts of data centre technology are becoming more compact and efficient.
New computer chips can perform more calculations in less physical space. Improved cooling systems can also help operators manage greater computing capacity within a given area.
Higher-voltage electrical designs can reduce some of the challenges associated with delivering large amounts of power to densely packed equipment.
Modular construction is another approach.
A modular data centre uses prefabricated sections or units that can be assembled and expanded as needed. Instead of building every part of the facility from scratch on site, operators can install standardised modules.
ASHRAE, an international engineering organisation concerned with heating, cooling and building systems, identifies modular and off-site construction as approaches that can help data centre operators add capacity more quickly and manage changing technology requirements. [3]
However, more compact equipment does not necessarily mean that future data centres will occupy less land overall.
If demand for AI continues growing, companies may use efficiency improvements to install even more computing capacity.
The result could be smaller equipment but larger campuses.
16. What happens inside an AI data centre when you ask a chatbot a question?
Understanding the physical structure becomes easier when we follow a simple AI request.
Imagine someone asks an AI chatbot to explain how a mobile phone works.
- The request travels through the internet: The user’s device sends the question to the AI service.
- The system receives the request: Computers inside the data centre receive and process the information.
- The AI model performs calculations: Specialised processors analyse the request and generate a response.
- Other equipment supports the process: Storage systems, networking equipment and power infrastructure help the computers operate.
- The cooling system removes heat: Cooling equipment manages the heat produced by the processors.
- The response returns to the user: The generated answer travels through the network and appears on the user’s screen.
The user may see the answer within seconds, but the process depends on a large physical system operating behind the scenes.
That is why AI, although it appears to exist mainly on the internet, has a substantial physical infrastructure. [1, 2]
17. Myths vs facts about AI data centres
| Claim | Status | What is actually true |
|---|---|---|
| Every AI data centre is the size of a city. | Myth | Facilities vary in size. Some are small server rooms, while others are large campuses. |
| AI data centres contain only computers. | Myth | They also need cooling, electrical systems, storage, networking, security and backup equipment. |
| The computers are the only reason AI data centres are large. | Myth | Supporting infrastructure, maintenance access and expansion requirements also take up space. |
| AI processors produce heat when they operate. | True | Their electrical activity generates heat that must be removed to protect the equipment. |
| Every AI data centre uses liquid cooling. | Myth | Cooling methods vary. Some use air cooling, others use liquid cooling, and some combine different approaches. |
| A larger data centre automatically produces a more intelligent AI model. | Myth | AI performance depends on model design, data, training and computing methods, not simply building size. |
| AI data centres need backup power systems. | True | Batteries and generators help maintain reliable operations during electricity interruptions. |
| AI server racks can require as much electricity as many households. | True | The IEA estimates that an advanced rack could have peak power demand equivalent to around 65 households by 2027. |
| All data centres are designed specifically for AI. | Myth | Many support cloud computing, websites, storage, business systems and other online services. |
| AI data centres may expand into multiple buildings. | True | Large operators can develop campuses to increase computing capacity and provide room for future expansion. |
Bottom line
AI data centres are so large because artificial intelligence depends on much more than a few powerful computers. These facilities must accommodate thousands of processors, connect them through high-speed networks, supply them with enormous amounts of electricity, remove the heat they generate and provide backup systems to keep operations running.
The rapid growth of AI is also changing how these buildings are designed. Modern facilities are being built to handle increasingly powerful processors and much higher computing demands, sometimes requiring specialised cooling and electrical infrastructure.
The next time an AI chatbot produces an answer in seconds, it is worth remembering that behind that small response is a large physical system of buildings, computers, cables, cooling equipment and power infrastructure working together.
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.
References
- [1] Key Questions on Energy and AI, International Energy Agency
- [2] Energy and AI: Energy demand from AI, International Energy Agency
- [3] Integrated Design Principles: AI Data Center Energy Performance Framework, ASHRAE
- [4] United States Data Center Energy Usage Report: 2025 Update, Lawrence Berkeley National Laboratory


