A data centre is a building, or part of one, built to house computer servers, storage, and networking equipment that run software and store data at scale. It supplies the power, cooling, connectivity, and physical security those machines need to run without interruption. When you stream a film, send an email, or query a search engine, the work happens inside a data centre somewhere, not on your own device.
Most people never see one. They sit on industrial land near power substations and fibre routes, often with no signage. Yet they underpin nearly every digital service you use, and the largest now cost billions of dollars to build.
This guide answers the common questions in order: what data centres do, how they work, how they are graded, how operators earn money, the main types, how many exist, and how AI is reshaping them. Each section is short enough to read on its own.
What is a data centre used for?
A data centre exists to keep computing running reliably. Organisations either build their own or rent space and power inside someone else's. The work falls into a handful of categories.
Hosting websites and applications. The servers behind a website, a banking app, or a logistics platform live in a data centre. Storing and backing up data. Photos, records, transaction logs, and archives are kept on disk arrays with redundant copies. Cloud computing. Providers such as the major cloud platforms rent out compute and storage on demand from large facilities.
Streaming and content delivery. Video and music services cache popular content close to users to cut delay. Enterprise systems. Email, payroll, customer databases, and internal tools run for staff across an organisation. AI training and inference. Modern facilities run dense clusters of accelerators to train and serve machine-learning models.
In short, anything that needs to compute, store, or move data at scale, around the clock, depends on a data centre.
How does a data centre work?
A data centre is engineered around four systems. Each has to work for the building to function.
Power. Racks of servers draw large, constant electrical loads. The facility connects to the grid, then adds uninterruptible power supplies and on-site generators so a grid fault does not bring servers down. Power capacity, measured in megawatts (MW), is the figure that defines a site's true size.
Cooling. Servers turn electricity into heat. Removing that heat is as demanding as supplying the power. Older sites use air cooling and chillers. Higher-density sites, especially AI ones, increasingly use liquid cooling that carries heat away directly from the chips.
Networking. Fibre connections link the building to carriers, internet exchanges, and other sites. Inside, switches and routers move data between servers at high speed. Good connectivity is why operators cluster in a few well-connected locations.
Security and resilience. Physical access is controlled with fencing, guards, biometrics, and cameras. Fire suppression, redundant equipment, and monitoring keep the site running through faults and maintenance.
Data centre tiers explained
Data centres are commonly graded by the resilience of their design. The most cited framework is the Uptime Institute tier classification, which runs from Tier I to Tier IV. The tiers describe how much redundancy a site builds into its power and cooling, and the indicative annual uptime each level targets.
Tier I is basic capacity with a single path for power and cooling and no redundancy, at roughly 99.671% annual uptime. Tier II adds redundant capacity components on a single distribution path, at roughly 99.741%. Tier III is concurrently maintainable, with multiple paths and one active, at roughly 99.982%. Tier IV is fault tolerant, with multiple active paths and full redundancy, at roughly 99.995%.
A higher tier means less downtime and more redundancy, at a higher build cost. Most enterprise and colocation demand sits at Tier III. Mission-critical and government workloads often require Tier IV.
How do data centres make money?
Data centres earn money in a few distinct ways, depending on who owns the building and who uses the space.
Colocation. The operator rents space, power, and cooling to many customers who bring their own servers. Customers pay for the rack space and the power they draw. Wholesale. Larger blocks of capacity, often whole halls or buildings, are leased to a single big tenant on long contracts. Hyperscale lease. A developer builds to the spec of a single cloud or technology giant and leases the whole site, frequently on 10 to 15 year terms.
Enterprise self-build. A company builds its own facility for internal use. Here the return is cost avoidance and control, not rent. REIT and infrastructure ownership. Many large operators are structured as real estate investment trusts or sit inside infrastructure funds, so investors earn from long-term, contracted rental income.
The common thread is recurring revenue tied to power capacity. A megawatt of leased, powered capacity is the underlying unit of value in the sector.
Types of data centre
Facilities differ by who owns them and what they are designed to do.
Enterprise. Owned and run by a single organisation for its own workloads. Colocation. Shared facilities where many customers rent space, common for businesses that want resilience without building their own. Hyperscale. Very large sites, often hundreds of megawatts, built by or for the largest cloud and platform operators.
Edge. Small facilities placed close to users to cut latency, used for content delivery, gaming, and increasingly AI inference. AI data centre. A facility built for the high power density and cooling that machine-learning workloads need, often with liquid cooling and racks drawing far more power than a traditional site.
How many data centres are there?
Counts vary by source and definition, since a small edge node and a 300 MW hyperscale campus are both technically data centres. The approximate figures below come from industry directories that track facilities by country, and they shift as new sites open.
United Kingdom: roughly 500, concentrated around West London and the Slough corridor. Australia: roughly 300, concentrated in Sydney, Melbourne, and Canberra. United States: the largest market by a wide margin at more than 5,000 facilities, with major clusters in Northern Virginia, Dallas, and Silicon Valley.
Counting by capacity matters more than counting buildings. A market with fewer, larger sites can hold far more compute than one with many small ones.
How data centres are changing for AI
AI has shifted what a data centre needs to be. Training and running large models concentrates enormous compute into dense clusters, which changes three things.
Power density. A traditional rack might draw 5 to 10 kW. AI racks now run far higher, often 50 to 150 kW or more, which means a single AI hall can demand as much power as an entire older facility.
Cooling. Air cooling struggles at these densities. Operators are moving to liquid cooling, including direct-to-chip and immersion, which raises water and engineering questions that planners now scrutinise closely.
Power and water supply. The binding constraint on new capacity is no longer land or money. It is whether the grid can deliver power, and when. In constrained markets, connection dates can slip by years, which is why the timing and certainty of power now decides where AI capacity can actually be built.
Frequently asked questions
What is the difference between a data centre and the cloud?
The cloud is a service: computing and storage you rent on demand over the internet. A data centre is the physical building where that service runs. Cloud providers operate large data centres, so the cloud lives inside data centres.
Why are data centres built where they are built?
They cluster where three things meet: available grid power, fibre connectivity, and suitable land with planning consent. Access to water for cooling and proximity to users also influence the choice.
What is an AI-ready data centre?
A facility built for high power density and intensive cooling, typically supporting 50 kW or more per rack with liquid cooling, the available power to feed it, and a confirmed connection date. Below roughly 30 to 40 kW per rack a site can usually rely on air cooling and is not considered AI-ready in the high-density sense.
How much power does a data centre use?
It varies widely. A small enterprise site may use a few megawatts; a hyperscale campus can require hundreds. Power use is the main measure of a facility's scale and its environmental footprint.
What does Tier III mean?
A Tier III data centre is concurrently maintainable, meaning equipment can be serviced without taking workloads offline, with redundant power and cooling paths. Under the Uptime Institute tier classification it targets roughly 99.982% annual uptime.