What Are AI Data Centres? Inside the Technology Powering AI Entertainment

An AI generated video, virtual performer or visual effect might appear on screen in seconds, but the work behind it can involve powerful computers operating inside huge facilities around the world. As artificial intelligence moves further into film, television, animation, music and digital storytelling, data centres are becoming an increasingly important part of the infrastructure behind entertainment.

For viewers and many creators, that infrastructure remains almost invisible. Data centres already consume a significant amount of electricity worldwide, and artificial intelligence is becoming an increasingly important part of that demand rather than being responsible for all of it.

What exactly is a data centre?

The UK Government defines a data centre as a physical facility where information technology equipment can be housed, connected and operated. Behind those computers is another layer of equipment providing electricity, cooling, ventilation, networking, security and protection against power failures.

In simpler terms, a data centre is a building designed to keep large numbers of computers running reliably around the clock. They existed long before the recent AI boom and already support streaming platforms, websites, cloud storage, banking, business software and many other online services.

What makes an AI data centre different?

An AI data centre is not an entirely different type of building. The main difference is the concentration and type of computing equipment inside it, along with the electricity and cooling needed to operate that equipment.

Modern artificial intelligence relies heavily on processors capable of performing enormous numbers of calculations very quickly. Graphics processing units, known as GPUs, and other specialised AI processors are particularly useful because they can carry out many calculations at the same time.

The amount of power being concentrated into these facilities is increasing rapidly. The International Energy Agency, known as the IEA, says that by 2027 a single rack of advanced servers could have a peak power demand equivalent to around 65 households. A server rack is essentially a large cabinet containing multiple pieces of computer equipment, so this gives a better idea of just how much power can be concentrated into a small space.

Training AI and using it are different

There are two important stages of AI computing that are often grouped together even though they create different demands.

Training is the process of developing an AI model. Large amounts of information are processed while the system adjusts how it works, and training a major model can require large groups of processors running for extended periods.

Inference happens when the trained model is actually used. If someone asks an AI system to generate dialogue, make an image, produce a voice or create a video, inference is the computing work required to deliver that result.

Training a large model can require substantial energy upfront, while inference can account for more total energy over time if a popular model is used millions or billions of times. This is one reason there is no useful single figure for the electricity consumed by “AI”.

How much electricity are data centres actually using?

The IEA’s April 2026 update estimates that data centres worldwide consumed around 485 terawatt hours of electricity in 2025. A terawatt hour, shortened to TWh, is one billion kilowatt hours, but that number becomes much easier to understand when compared with something familiar.

UK Government figures show that total UK electricity demand during 2025 was around 320 TWh. That means data centres worldwide used roughly one and a half times as much electricity as the entire UK during the same year.

The IEA projects global data centre electricity use could rise to around 950 TWh by 2030. At the UK’s 2025 level, that would be almost three times the electricity used by the whole country in a year.

Those figures cover all data centres, not artificial intelligence alone. They include the infrastructure supporting streaming, cloud services, websites and many other digital activities, but the IEA expects electricity consumption from data centres focused on AI to triple between 2025 and 2030.

Why is cooling so important?

Powerful processors generate heat while they work, and concentrating large numbers of them inside one facility creates a significant cooling challenge. Keeping the equipment at safe temperatures requires additional technology and electricity.

The UK Government’s April 2026 Compute Evidence Annex says cooling can account for up to 40% of a data centre’s total energy consumption. AI can make that challenge greater because powerful processors may operate intensively for long periods and be packed closely together.

Cooling systems can use air, water or different forms of liquid cooling depending on the facility. Water requirements also vary considerably according to the technology being used, the local climate and whether water is reused within a closed system.

What does this mean for entertainment?

The connection with entertainment is becoming increasingly direct. AI is being explored for visual effects, animation, dubbing, voice generation, production planning and digital storytelling, while independent creators are also producing complete images, characters and videos using generative tools.

Research published in June 2025 by the British Film Institute, known as the BFI, and the UK’s CoSTAR creative research network examined the growing use of generative AI across the screen industries. It highlighted opportunities including faster workflows and greater access to creative tools, while also identifying high energy consumption and carbon emissions as issues that need to be considered.

How AI is used can make a difference too. Generating numerous versions of the same visual effects shot, for example, can mean repeatedly asking an AI model to process new requests before a final version is chosen, increasing the total computing required.

That does not make AI an automatically unsustainable production tool. More efficient models and processors can reduce the energy needed for individual tasks, while AI could potentially make some production processes quicker or reduce the need for other resources.

Are AI data centres bad for the environment?

There is no simple yes or no answer because the environmental impact varies according to where a facility is built, where its electricity comes from, how efficient its equipment is and how it is cooled. A data centre supplied largely by lower carbon electricity can have a very different carbon footprint from one relying heavily on fossil fuels.

The challenge is that efficiency is improving at the same time as demand is growing. AI video, larger models and wider adoption across entertainment and other industries could push overall electricity use higher even if each individual task becomes more efficient.

The infrastructure audiences rarely see

One of the most striking things about AI entertainment is the difference between what appears on screen and what is required behind it. A short video generated on a phone can depend on specialist processors, industrial scale electricity supplies, cooling equipment and data centre infrastructure located hundreds or thousands of miles away.

That does not mean every AI generated image, song or video should be viewed as an environmental problem. It does mean that as artificial intelligence becomes a bigger part of film, television, music and digital storytelling, the physical infrastructure behind those creative tools deserves to be understood too.

For more on how artificial intelligence is changing film, television, music and digital storytelling, explore SoapLandTV’s AI & Entertainment page. You can also find our latest explainers, industry coverage and stories following the growth of AI created entertainment.

Sources: International Energy Agency, April 2026; UK Government Energy Trends, March 2026; UK Government Compute Evidence Annex, April 2026; British Film Institute and CoSTAR, June 2025.