What is a data center?

A data center is a building, or group of buildings, that houses the computers, storage and networking equipment used to run digital services, together with the power and cooling systems that keep that equipment running around the clock.

Every time you stream a video, send an email, use a banking app or ask an AI assistant a question, the work is done on servers in a data center somewhere. Facilities range from a single room in an office to hyperscale campuses covering hundreds of thousands of square metres.

How a data center works

A data center turns electricity into computing. Power arrives from the grid through substations and transformers, is conditioned and backed up by uninterruptible power supplies (UPS) and generators, and is distributed to racks of servers. The servers process and store data, and high-capacity network links connect them to each other and to the internet. Almost all of the electricity consumed ends up as heat, so cooling systems continuously remove it.

Main components

• Servers: the computers that run applications, including CPUs and, increasingly, GPUs and other AI accelerators.
• Storage: disk and flash systems that hold data.
• Networking: switches, routers and fibre that move data inside the building and out to the internet.
• Power infrastructure: grid connection, transformers, UPS batteries, backup generators and power distribution units.
• Cooling: air handling, chillers, cooling towers and, in newer AI facilities, liquid cooling delivered directly to chips.
• Physical security and monitoring: access control, fire suppression and building management systems.

Types of data centers

• Enterprise data centers are owned and operated by a single organisation for its own use.
• Colocation data centers rent space, power and cooling to many customers, who install their own equipment.
• Hyperscale data centers are very large campuses run by cloud and internet companies such as Amazon, Microsoft, Google and Meta.
• Edge data centers are smaller sites placed close to users to reduce latency.
• AI data centers are designed specifically for dense clusters of GPUs used to train and run AI models.

What makes an AI data center different

AI workloads are run on thousands of GPUs, such as NVIDIA's H100 and H200 or Blackwell-generation B200 and GB200 systems, linked by very fast networks. A rack of these systems can draw more than 100 kilowatts, compared with roughly 5 to 15 kilowatts for a typical enterprise rack. That density changes almost every design decision, from floor loading to cooling. Read more in our guide What is an AI data center?.

Power and cooling

Power is now the main constraint on building new data centers. Large AI campuses are planned in hundreds of megawatts, and some in gigawatts, which requires new grid connections, long-term energy contracts and, increasingly, on-site generation. Efficiency is often measured by power usage effectiveness (PUE): the ratio of total facility power to the power used by IT equipment, where 1.0 would be perfect. See How AI data centers use electricity and our energy coverage.

Cooling has moved from air towards liquid. Direct-to-chip liquid cooling and, in some cases, immersion cooling are used to handle the heat produced by modern accelerators.

Where data centers are built

Operators look for reliable and affordable power, fibre connectivity, available land, a suitable climate, supportive regulation and access to skilled staff. Established markets include Northern Virginia, Dallas, Phoenix, London, Frankfurt, Amsterdam and Dublin; see the largest data center markets in Europe.

Follow data center news

GPU Data Hub tracks new campuses, capacity, financing and power deals every day. Browse the AI data center directory, the latest data center news, deals and funding and the Daily Brief.