The cloud is a building
When people say their photos or documents are “in the cloud,” the data is stored on computers in a real building, usually a purpose-built facility with controlled power, cooling and security. These buildings are called data centers. They range from a small room in an office to enormous campuses with many halls of equipment, and they run the services behind web sites, streaming, online banking, messaging and a growing share of artificial intelligence work.
The word “cloud” is mostly a way of describing how computing is rented and managed: users get capacity on demand without owning the machines. The business side of that is covered in discussions of cloud computing, while this article is about the physical side.
What is inside
Servers. The core equipment is a large number of server computers, typically mounted in tall metal frames called racks. A server is much like an ordinary computer without a screen, built to run continuously. Racks are arranged in long rows.
Storage. Some machines hold large amounts of data on drives. Data is commonly copied to more than one device, and often more than one building, so that a failed drive or even a failed site does not mean lost information.
Networking. Switches and routers connect the servers to each other and to the outside world. Inside a large facility the internal traffic between servers can be considerable, and the building usually connects to several outside network providers so that one failure does not cut it off. Those outside links ultimately join the global network of fiber described in explanations of how the internet works.
Power equipment. Electricity enters from the grid and passes through transformers and distribution equipment. Batteries or similar systems, often called uninterruptible power supplies, bridge short interruptions, and generators typically provide longer backup. The aim is for the computers to keep running through a power failure.
Cooling. Almost all the electricity that computers use ends up as heat, which must be removed or the equipment will overheat. Cooling is therefore a major part of the design.
Security and monitoring. Facilities usually control physical access carefully and monitor temperature, power and equipment health around the clock.
How cooling works
Common approaches include moving cooled air through aisles of racks, often arranged so that cold air and hot exhaust air are kept apart. Chilled water systems carry heat away from the room. In cooler climates, outside air can sometimes do part of the work. Some newer high-power equipment is cooled by circulating liquid close to the chips, because air alone may struggle with very dense racks. Which approach suits a site depends on climate, equipment and cost, and designs keep changing.
Why location matters
Operators weigh several things when choosing a site: the price and reliability of electricity, access to fiber, the climate, land, local regulation and the risk of natural hazards. Delay also matters. A service that must respond quickly to users may need servers relatively near them, which is one motivation for edge computing, where smaller facilities sit closer to where people live. Other workloads, such as storing backups, are less sensitive to distance and can sit wherever running costs are lower.
The energy question
Data centers use a lot of electricity in total, and demand has been a topic of public discussion, particularly as interest in AI has grown. Exact figures differ between sources and depend on what is counted, and forecasts are uncertain, so it is wise to treat any single headline number with caution.
Some basic points are widely accepted:
- Efficiency has improved. Computers do more work per unit of energy than they once did, and facilities waste less on overhead such as cooling. A common measure, power usage effectiveness, compares the total energy a facility uses with the energy reaching the computing equipment. A value closer to one means less overhead.
- Growth can outpace efficiency. Even if each task needs less energy, a rapid rise in the number of tasks can still increase total demand.
- Local effects can be large. A single large campus can be a significant new load for a regional grid, which raises questions about whether the grid and generation can keep up and who pays for upgrades.
- The source of electricity matters. Whether a facility’s power comes largely from fossil fuels or from low-carbon sources changes its climate impact a great deal. Many operators buy renewable power, though matching supply to round-the-clock demand is harder than it sounds, which ties into why intermittent renewables benefit from battery storage and other flexibility.
Water
Some cooling systems evaporate water to shed heat, and so can use substantial amounts of it. Others use little or none but may need more electricity instead. This trade-off is a concern mainly in water-stressed regions, and communities and regulators in some places have asked operators to disclose and reduce use. Quantities differ widely from site to site.
Reusing heat and other ideas
Because data centers produce a steady stream of warm air or water, some projects have supplied that heat to nearby buildings through district heating, where the local infrastructure makes this practical. Other ideas include locating facilities in cooler regions, using waste heat for greenhouses or pools, and scheduling flexible workloads for times when clean power is plentiful. Each has limits, and how widely they work is still being established.
Because big facilities are such concentrated loads, they are also discussed alongside smart grids, which aim to make demand more responsive to supply.
Reliability and tiers
Because so many services depend on them, designers add redundancy: duplicate power feeds, spare cooling units and multiple network paths. More redundancy typically means higher cost, so critical services pay for more of it than casual ones. Even so, outages happen, from power faults, cooling failures, software errors or human mistakes.
A fair summary
Data centers are neither mysterious nor trivial. They are industrial buildings that turn electricity into computation and heat, and their value to everyday life is large. The open questions are mostly about scale: how fast demand will grow, how clean the supply will be, and how to share the costs and benefits locally.