A data center is a dedicated facility that houses the servers, storage systems, and networking equipment that run applications, websites, and data storage for organizations — along with the power, cooling, and physical security needed to keep that equipment running without interruption. In practice, almost every app, website, and AI model you use is served from inside one of these buildings. Globally, the facilities run by the largest cloud providers alone now number in the thousands, and demand for new capacity is growing faster than at any point in the industry’s history.
What Is a Data Center, Exactly?
At its core, a data center is a building (or part of one) engineered around three jobs: compute, storage, and networking. Racks of servers process and store data; switches and routers move that data in and out over fiber connections; and a layer of supporting infrastructure — uninterruptible power supplies (UPS), backup generators, computer room air conditioning (CRAC) or liquid cooling, fire suppression, and physical security — keeps all of it running 24/7/365.
What separates a data center from a server closet in an office is redundancy. A real data center is built so that a single failed component — a power feed, a chiller, a network switch — doesn’t take the whole facility down. That design discipline is what the tier classification system (more on that below) is built to measure.
What Does a Data Center Actually Do?
A data center’s job is to keep IT equipment powered, cool, connected, and secure so the applications running on it stay available. Four functions make that possible:
- Compute and storage: Servers run the applications; storage arrays (disk and flash) hold the data those applications read and write.
- Networking: Switches, routers, and fiber links connect servers to each other and to the internet or a private network.
- Power infrastructure: Utility feeds, UPS systems, and backup generators keep equipment running through outages — this is usually the single biggest capital cost in the building.
- Cooling and environmental control: Air- or liquid-based cooling systems remove the heat that dense server racks generate, which is increasingly the limiting factor as AI hardware gets denser.
Types of Data Centers
Not all data centers are built or owned the same way. The four common categories are enterprise, colocation, hyperscale, and edge. The table below compares them.
| Type | Who Owns/Operates It | Typical Scale | Best Suited For |
|---|---|---|---|
| Enterprise | A single company, for its own use | A few hundred kW to a few MW | Organizations with specific compliance, latency, or legacy-system needs |
| Colocation | A third-party operator; tenants rent space | Shared facility, tenant suites from a few kW to several MW | Companies that want to outsource the building but keep their own hardware |
| Hyperscale | Cloud/AI giants (AWS, Microsoft, Google, Meta, Oracle, etc.) | Tens to hundreds of MW per campus | Cloud computing, SaaS platforms, and large-scale AI training/inference |
| Edge | Telecoms, CDNs, or cloud providers | Small footprint, close to end users | Low-latency applications: content delivery, IoT, real-time AI inference |
The hyperscale category has grown the fastest. Synergy Research Group counted 1,360 large data centers operated by hyperscale companies at the end of Q4 2025, with almost 800 more in the known pipeline. Synergy’s data shows hyperscale operators already hold 48% of worldwide data center capacity, up from a minority share in 2018 when on-premise facilities still held 56%; colocation holds roughly 20% today. Synergy projects hyperscale operators will control 67% of all data center capacity by 2031, with on-premise enterprise facilities shrinking to about 19%. If you’re deciding whether to build your own facility, buy capacity, or lease colocation space, that shift matters: it’s a big part of why colocation pricing and availability have tightened across most major markets.
Data Center Tiers: How Reliability Is Classified
The most common framework for rating a data center’s resilience comes from the Uptime Institute’s Tier Classification System. It doesn’t assign a guaranteed uptime percentage to each tier — Uptime Institute removed “expected downtime per year” references from the standard back in 2009 — but it does define increasing levels of redundancy:
- Tier I (Basic Capacity): A dedicated IT space with a UPS, dedicated cooling, and a backup generator, but no redundant components.
- Tier II (Redundant Capacity): Adds redundant power and cooling components (extra UPS modules, chillers, pumps, generators), giving a margin against equipment failure.
- Tier III (Concurrently Maintainable): Adds a second, independent power and cooling delivery path so any component can be taken offline for maintenance without affecting IT operations.
- Tier IV (Fault Tolerant): Builds on Tier III so that even an unplanned failure or distribution-path interruption is contained before it reaches IT equipment.
Most colocation and hyperscale facilities built in the last decade target Tier III or above. For a deeper look at what actually changes between the two most commonly specified tiers, see our Tier III vs. Tier IV comparison.
How Much Power Does a Data Center Use?
Power is the resource that increasingly decides where a data center can be built at all. According to the International Energy Agency’s “Energy and AI” report, data centers worldwide consumed roughly 415 TWh of electricity in 2024 — about 1.5% of global electricity consumption. The IEA’s base case projects that figure will more than double to around 945 TWh by 2030, just under 3% of projected global electricity consumption that year. A single hyperscale campus can now draw as much power as a mid-sized city, which is why utility interconnection queues, not construction schedules, are the real bottleneck in many markets today.
Need Expert Guidance?
Talk to a Data Center Expert
21+ years of hands-on experience in data center design, operations & infrastructure. Book a quick discovery call to discuss your project.
📞 Book a Discovery CallPower consumption also determines a facility’s efficiency rating. Most operators track power usage effectiveness (PUE) — the ratio of total facility power to the power actually delivered to IT equipment — as the standard benchmark for how much energy goes to cooling and overhead versus useful compute. A PUE close to 1.0 means almost all power reaches the servers; older air-cooled enterprise facilities often run well above that, which is one reason enterprises have been moving workloads into more efficient colocation and hyperscale facilities rather than operating their own aging rooms.
What’s Driving Data Center Growth Right Now
AI training and inference are the primary reason capacity demand has accelerated so sharply. McKinsey’s Global Institute estimates the world will need to invest $6.7 trillion in data centers by 2030 to keep pace with compute demand, with AI-focused facilities accounting for roughly $5.2 trillion of that total and non-AI workloads the remaining $1.5 trillion. McKinsey puts AI-related data center capacity demand at 156 GW by 2030, with 125 incremental gigawatts added between 2025 and 2030 in its base case — a range that spans 78 GW in a constrained scenario to 205 GW in an accelerated one. For buyers, the practical takeaway is that the gap between planned and available capacity is widening, which is pushing lead times and pricing up across colocation and build-to-suit deals alike.
It’s worth adding some nuance to the growth narrative, though. Not every forecast assumes unconstrained demand: McKinsey’s own range runs from a 78 GW constrained scenario to a 205 GW accelerated one, a sign that analysts themselves disagree on how much of today’s AI buildout will actually be utilized versus how much reflects speculative pre-leasing. Efficiency gains partly offset raw growth too — newer liquid-cooled, higher-density racks do more useful compute per megawatt than the air-cooled halls built a decade ago, which is part of why the IEA’s 2030 electricity projection (945 TWh) is less than triple 2024’s 415 TWh despite far larger compute workloads. Buyers evaluating a long-term lease or build decision should treat any single growth number as a scenario, not a certainty, and stress-test commitments against the slower end of these ranges.
Frequently Asked Questions
What is the simplest definition of a data center?
A data center is a physical facility that houses servers, storage, and networking equipment, along with the power and cooling systems needed to keep that equipment running continuously.
What is the difference between a data center and the cloud?
The cloud is a service delivered from data centers. Cloud providers like AWS, Microsoft Azure, and Google Cloud operate hyperscale data centers and rent out compute, storage, and networking capacity to customers rather than selling physical space.
What is the difference between a data center and colocation?
A data center is the facility itself. Colocation is a business model where a third-party operator owns and runs the building, and customers rent rack space, power, and cooling for their own servers inside it.
How big is a typical data center?
It varies enormously. A small enterprise data center might be a few hundred kilowatts of IT load in a few thousand square feet, while a hyperscale campus can exceed 100 megawatts across more than a million square feet.
Why are data centers suddenly in the news so often?
AI workloads require far more power-dense compute than previous cloud applications, which is straining electricity grids, water supplies, and construction supply chains in major markets — and raising the capital needed to build new capacity, per McKinsey’s $6.7 trillion 2030 estimate above.
If you’re evaluating how much data center capacity your organization actually needs, and whether to build, buy, or lease it, start with our data center buyer’s toolkit for a structured way to work through the decision.
Sources
- International Energy Agency, “Energy and AI” — Energy demand from AI
- Synergy Research Group — Hyperscale Operators to Account for 67% of All Data Center Capacity by 2031
- McKinsey Global Institute — The Cost of Compute: A $7 Trillion Race to Scale Data Centers
- Uptime Institute Journal — Explaining the Uptime Institute’s Tier Classification System
Written by
Raajeev Ratra
Data Center Infrastructure Expert | 21+ Years in DC Design, Operations & Project Management
Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.