As artificial intelligence becomes more capable and widespread, it comes with a hidden cost that is drawing increasing attention: the enormous amount of energy required to power the data centres behind it. This guide explains why AI uses so much energy, what impact that has, and what is being done to make the computing infrastructure that powers AI more sustainable.
What data centres are
Data centres are large facilities filled with computer servers that store data and run the computations behind online services, including AI. When you use AI or many other digital tools, powerful hardware in these centres is doing the work. Understanding that data centres are the physical backbone of the digital and AI world is the starting point for grasping why their energy use matters so much.
Why AI is energy-intensive
Training and running advanced AI models requires enormous amounts of computation, which consumes significant electricity. The powerful hardware involved draws a lot of power and generates heat that must be managed. As AI models grow larger and are used more widely, their energy demands increase. Understanding that AI's capabilities come with substantial computational and energy costs explains why this issue is growing.
The cooling challenge
The hardware in data centres generates heat, and keeping it cool requires additional energy, adding to overall consumption. Cooling is a major part of a data centre's energy use. This means the energy footprint of AI includes not just running the computations but also keeping the equipment at safe temperatures, a factor often overlooked when considering the true cost of AI.
The environmental impact
The electricity powering data centres often comes partly from sources that generate emissions, so heavy energy use can contribute to environmental impact. Data centres also require resources like water for cooling in some cases. As AI expands, its cumulative environmental footprint becomes more significant, prompting growing concern about balancing technological progress with sustainability.
Efforts to reduce the impact
There are many efforts to make data centres and AI more sustainable, including using more efficient hardware, improving cooling, running facilities on cleaner energy, and designing more efficient AI models. These efforts aim to reduce the energy and environmental cost of computing. Understanding them shows that the industry is aware of the challenge and actively working to address it.
Why this matters
As AI becomes woven into more of daily life, its energy use is an important part of understanding its true cost. Awareness of this hidden cost encourages more responsible development and use of AI. Understanding data centre energy use helps you see the fuller picture of AI, appreciating both its benefits and the real-world resources required to deliver them.
Training and inference are different costs
Discussions of AI energy use often blur two very different things. Training is a one-off, concentrated expenditure: a large model is trained once, at considerable cost, over weeks on many processors. Inference is the ongoing cost of actually using the model, paid in small amounts every time someone submits a query.
The distinction matters because the two scale differently. Training cost is fixed regardless of how many people eventually use the model, so it is amortised across all subsequent use. Inference cost grows directly with adoption. A widely used model can therefore consume far more total energy serving requests over its lifetime than it ever did during training, even though the training figure is the one that makes headlines.
Where the electricity actually goes
Inside a data centre, not all power reaches the computation. A meaningful share goes to cooling, because processors convert essentially all the energy they draw into heat that must then be removed. Additional power is lost in conversion and distribution, and some supports networking, storage and lighting.
The industry measures this with power usage effectiveness, the ratio of total facility power to the power delivered to computing equipment. A ratio of 1.0 would mean every watt reaches the servers, which is physically unattainable. Well-run modern facilities operate close to the low 1.1s; older or poorly sited ones are considerably worse. The gap between those figures represents real energy spent on overhead rather than work.
Water is the less visible constraint
Cooling frequently uses evaporative systems, which consume water rather than merely circulating it. In temperate regions with plentiful supply this attracts little attention. In water-stressed areas it creates direct competition with agriculture and domestic use, and has become a genuine point of local conflict around several large facilities.
Air cooling avoids the water consumption but uses more electricity, which shifts the burden rather than removing it. There is no configuration with no cost; there are trade-offs between electricity, water and capital expenditure, weighted differently depending on where the facility sits.
What actually reduces the footprint
Location is the largest single lever, because the same computation produces very different emissions depending on the grid supplying it. A data centre on a grid dominated by hydroelectric or nuclear generation emits a fraction of one running on coal, with no change to the hardware at all.
Timing helps as well. Workloads that are not urgent, such as training runs and batch processing, can be scheduled for periods when renewable generation is abundant, a practice known as carbon-aware scheduling. Beyond that, smaller specialised models often deliver adequate results for a given task at a small fraction of the energy of a large general-purpose one, which makes model selection an efficiency decision rather than purely a capability decision.
Common questions
Why does AI use so much energy?
Training and running advanced AI models requires enormous computation on powerful hardware, which consumes significant electricity and generates heat.
What are data centres?
Data centres are large facilities of computer servers that store data and run the computations behind online services and AI.
How do data centres affect the environment?
Their heavy electricity use, often partly from emission-generating sources, and resource needs like cooling contribute to an environmental footprint.
What is being done to reduce AI's energy impact?
Efforts include more efficient hardware and cooling, running facilities on cleaner energy, and designing more efficient AI models.