A data center is a building, or a dedicated part of one, built to house computing equipment such as servers, storage and network gear. What separates it from an office server closet is the supporting infrastructure: utility power backed by batteries and usually generators, cooling sized for the equipment’s heat, physical security, fire protection and connections to one or more networks. Data centers range from a single room in a corporate campus to large multi-tenant buildings run by colocation and cloud providers.
At a glance
- A data center is defined by its supporting infrastructure (power, cooling, security, connectivity), not by who owns it.
- Enterprises run their own, rent space in colocation facilities, or use capacity in cloud providers’ data centers; many use a mix.
- Redundancy levels vary widely, from a single power and cooling path to designs meant to keep running through equipment failures and maintenance.
- Location, available networks and power density per rack often matter as much to buyers as floor space.
What problem it solves
Servers and network equipment need steady power, a controlled temperature and protection from people who should not touch them. An office building is designed for people, not equipment. Its power may fail during storms, its air conditioning shuts off on weekends, and its network options are whatever carriers happen to serve the building. Equipment kept in a closet tends to fail at the worst times.
A data center concentrates that equipment somewhere the infrastructure is engineered for it. Backup power covers utility outages, cooling is sized and monitored for the load, access is controlled and logged, and multiple network providers are often available on site. For a business, the question is less “do we need a data center” than “whose data center”: its own, a colocation provider’s, or a cloud provider’s.
How it works
Power. Utility feeds come into the building and pass through switchgear, uninterruptible power supplies (UPS) and power distribution units to each rack. Batteries in the UPS carry the load for the short time it takes standby generators to start. How many of these components are duplicated, and whether there are one or two independent paths to each rack, determines how the facility handles failures and maintenance. Designs are often described in terms such as N+1 redundancy or 2N.
Cooling. Equipment turns nearly all of the power it draws into heat. Cooling systems (air handlers, chillers, and in some facilities liquid cooling) remove it. Hot-aisle and cold-aisle layouts keep exhaust air from mixing with intake air. A facility’s efficiency is commonly expressed as power usage effectiveness (PUE).
Space. Equipment sits in racks or cabinets on a raised floor or slab. In multi-tenant buildings, customers rent cabinets, caged areas or private suites.
Connectivity. Carriers bring fiber into the building, often to a meet-me room, and customers connect to them and to each other with cross-connects. A carrier-neutral data center lets many unaffiliated networks operate on site, which gives tenants a choice of providers.
Operations. Staff monitor the infrastructure, handle maintenance and, in colocation facilities, perform hands-on tasks for customers. Designs are often described using a tier classification, which rates how well the power and cooling are designed to tolerate maintenance and failures.
Scale varies. Small edge computing sites sit close to users or devices; hyperscalers operate very large campuses to run their public clouds.
When it matters for buyers
- Moving offices. If servers live in the current office, a move is a natural time to decide whether they should go to colocation, the cloud or a new server room.
- Inheriting infrastructure. A new IT leader or an acquisition often brings a server room nobody has assessed. Check its power, cooling, age and single points of failure.
- Outages. Repeated power or cooling events in an office server room are a sign the equipment has outgrown the space.
- Hardware refresh. Replacing servers is the moment to compare owning hardware in colocation, renting bare metal or moving workloads to the cloud.
- Compliance or customer requirements. Auditors and customers may ask where data is hosted and what controls apply to the physical facility.
To compare facilities and providers, see our Colocation solution page.
Questions to ask vendors
- What redundancy does the facility’s power and cooling design have, and has that design been independently certified?
- How long can the site run on generator power, and how is fuel resupply arranged?
- What power density per rack can you support today, and how is power billed?
- Which carriers and cloud on-ramps are present in the building, and what do cross-connects cost?
- What compliance reports (for example SOC 2 or ISO/IEC 27001) cover the facility?
- What are the access procedures, and how quickly can you perform on-site tasks for us?
- What does the service level agreement cover (power, cooling, network), and what are the remedies?
- What is the site’s history of outages, and how were they handled?
How it differs from colocation
A data center is the facility itself. Colocation is a service sold inside one: the provider owns or operates the building and its infrastructure, and you rent space, power and connectivity for equipment you own. Every colocation offering lives in a data center, but not every data center offers colocation; many are private enterprise facilities or are used only by a cloud provider for its own services.