Data Center Glossary — 50 Terms Explained by Operators

Power

IT Load: The power actually consumed by servers, storage, and networking — the "useful" work the building exists to deliver. Data centers are sized, priced, and sold in megawatts of IT load, not square feet: a "10 MW facility" delivers 10,000 kW at the rack face.
PUE (Power Usage Effectiveness): Total facility power ÷ IT load. A perfect score is 1.0. Modern hyperscale facilities achieve 1.08–1.2; the industry average sits around 1.5. On a 100 MW campus, the gap between those two numbers is millions of dollars a year in power cost.
UPS (Uninterruptible Power Supply): The battery-backed system that carries the full IT load during the 10–15 seconds it takes generators to start — typically with 5–15 minutes of battery runtime. Double-conversion is the dominant topology; modern units run 96–98% efficient at load.
PDU (Power Distribution Unit): Makes the final voltage step-down — 480 V to 415/240 V — at the edge of the data hall and feeds the rack busways or whips. Per-circuit metering here is what makes PUE measurable. Dual PDUs on independent A/B paths feed dual-corded servers.
ATS (Automatic Transfer Switch): Detects loss of utility power and transfers the load to generators, completing the swap in roughly 10 seconds. The UPS bridges that window so the servers never notice.
Backup Generator: Diesel units — typically 2.5–3 MW each — deployed in N+1 or 2N configurations, with 48–96 hours of fuel on site and priority refueling contracts. Permitting them (air quality) can be its own battle.
Transformer: Steps grid voltage down at each stage of the power chain. Large substation transformers are running 2–4 year lead times in 2025–2026 — the single longest-lead item on a project, and the reason developers place orders before design is finished.
Switchgear: Assemblies of circuit breakers, bus bars, and protective relays that route power through the facility and define the A/B distribution paths redundancy depends on. Medium-voltage switchgear lead times have stretched to 40–52 weeks.
Substation: Steps transmission voltage (typically 115–230 kV) down to medium voltage (~34.5 kV) for the campus. A large customer substation runs $15–50M and often paces the whole project.
N+1 / 2N Redundancy: N is the capacity required to run the load. N+1 adds one shared spare unit; 2N means two complete, independent systems, either able to carry the whole facility alone. Tier IV facilities require 2N-style fault tolerance.
Interconnection Queue: The utility's waitlist for new grid connections. System impact studies take 6–18 months (paid by the developer), and constrained markets run 3–7 years to energization — usually the true critical path of a project.
Busway (Bus Duct): Rigid metal enclosures carrying copper or aluminum bus bars for high-current distribution — preferred over cable in large data halls for reliability and easier maintenance.

Cooling

CRAC / CRAH: Computer Room Air Conditioner (its own refrigerant circuit) and Computer Room Air Handler (a coil fed by chilled water). The workhorse units that condition data hall air; large facilities overwhelmingly use CRAHs on a central chilled-water plant.
Chilled Water Loop: The piped loop carrying cool water from the chiller plant to the air handlers. Supply temperatures have risen from ~45°F historically toward 60°F+ in modern designs — warmer water means far less chiller energy.
Chiller: The refrigeration machine at the heart of the chilled-water plant and one of the largest mechanical loads in the building — a big share of the gap between PUE 1.5 and PUE 1.2.
Cooling Tower: Rejects the facility's heat to the atmosphere by evaporating water — which is why cooling design drives a data center's water footprint and, increasingly, its permitting battles.
Hot Aisle / Cold Aisle Containment: Arranging racks so cold supply air and hot exhaust never mix, with physical barriers sealing the aisles. One of the highest-ROI efficiency measures in any air-cooled hall.
Direct-to-Chip Liquid Cooling: Cold plates mounted on CPUs and GPUs carry heat away in liquid, which moves heat orders of magnitude better than air. Effectively required as racks pass ~50 kW — and AI training racks now run 80–150 kW+.
Rear-Door Heat Exchanger: A water-fed coil mounted on the back of a rack that neutralizes exhaust heat before it enters the room — a common bridge step between traditional air cooling and full liquid cooling.
Immersion Cooling: Submerging entire servers in non-conductive dielectric fluid. The densest cooling method available, still a niche deployment compared with direct-to-chip.
Free Cooling / Economizer: Using cool outside air or water instead of running compressors. The reason cool, dry climates post the industry's lowest PUEs.
WUE (Water Usage Effectiveness): Liters of water consumed per kWh of IT energy — the water counterpart to PUE, now scrutinized by communities and regulators in dry markets.
Delta T (ΔT): The temperature rise of air or water across the IT load. A higher ΔT moves the same heat with less airflow — less fan energy, smaller ducts, better economics.

Network

Carrier Hotel: A highly interconnected building where dozens or hundreds of networks physically meet. The most network-dense real estate in any metro — and nearly impossible to replicate, because the value is the tenants already inside.
Meet-Me Room (MMR): The secure room inside a facility where tenants and carriers interconnect. Control of the MMR — and the fees charged there — is a meaningful economic lever for the operator.
Cross-Connect: A physical cable connecting two parties within a facility. Individually small monthly fees at very high margin — and because ripping out hundreds of cross-connects is painful, they make interconnection-rich buildings extremely sticky.
Dark Fiber: Installed but unlit fiber leased to a customer who runs their own optical equipment over it — maximum control and capacity, used heavily by hyperscalers between campuses.
Latency: Signal delay. Light in fiber covers roughly 100 km per 0.5 ms, so distance sets a hard physical floor — the physics behind edge siting, market clustering, and why some workloads can't move to cheap remote power.
Peering / Internet Exchange: Networks exchanging traffic directly with one another — usually settlement-free — at an exchange point, instead of paying a transit provider to carry it.
Fiber Diversity: Multiple physically separated fiber routes and building entry points, so no single construction backhoe can isolate the facility. A standard item on every tenant's diligence checklist.
Edge Data Center: A small facility placed close to end users to minimize latency — content delivery, gaming, and local AI inference are the classic workloads.

Real Estate & Deals

Hyperscaler: The largest cloud and AI platforms — AWS, Microsoft, Google, Meta — operating fleets of massive facilities. They build campuses of hundreds of megawatts, lease wholesale capacity in huge blocks, and run the industry's best efficiency at PUE 1.08–1.2.
Colocation: Leasing space, power, and cooling in a shared facility. Retail colo sells by the cabinet or cage; wholesale colo sells whole data halls or megawatts to large tenants under long-term leases.
Build-to-Suit: A facility developed to a single tenant's specification, with the lease signed before construction starts. Lower risk and lower return than speculative development — the dominant model for hyperscaler capacity.
Powered Shell: A completed building with utility power delivered but interiors unfinished; the tenant installs its own electrical and mechanical fit-out. A distinct leasable product with its own pricing.
Powered Land: Entitled land with secured grid capacity. Because power — not land — is the scarce resource, powered land trades at a large premium over ordinary industrial acreage.
NNN (Triple-Net) Lease: The tenant pays property taxes, insurance, and maintenance on top of base rent, leaving the landlord with a clean income stream. The standard structure for wholesale and hyperscale leases, commonly 10–15+ year terms.
Escalations: Contractual annual rent increases, typically 2–3% per year in data center leases — a major driver of long-term deal value at underwriting.
Power Pass-Through: A lease mechanism where the tenant pays actual utility costs directly or at cost. It insulates the landlord from electricity price risk — and makes the tenant care intensely about the facility's PUE.
$/kW/Month: The pricing unit of wholesale leases: capacity is priced per kilowatt of critical IT load per month, not per square foot. The clearest sign the asset class sells power delivery, not floor area.
Cap Rate: Net operating income ÷ asset value — the yield metric for pricing stabilized assets. Long NNN leases to investment-grade hyperscale tenants command the tightest cap rates in the sector.
Entitlement: The legal right to build: zoning, site-plan approval, environmental review, and permits. A fully entitled, powered site can be worth multiples of the same land without approvals.

Operations & Reliability

Uptime Tiers (I–IV): The Uptime Institute's reliability classification. Tier I/II are basic; Tier III is concurrently maintainable; Tier IV adds fault tolerance — 2N-style systems where any single failure is absorbed without dropping load.
Concurrent Maintainability: The ability to take any single component offline for service without interrupting the IT load — the defining requirement of Tier III and a baseline expectation of serious tenants.
Five Nines: 99.999% availability — about 5.26 minutes of downtime per year. Each added "nine" multiplies the redundancy (and capital) required to promise it.
SLA (Service Level Agreement): The contractual uptime guarantee to tenants, with financial credits owed when it's breached. The SLA is where design redundancy becomes a legal and financial commitment.
Commissioning (Cx) Levels 1–5: The staged testing that proves the facility works: factory witness tests (L1), delivery inspection (L2), component startup (L3), system functional testing (L4), and full-building integrated systems testing under simulated load (L5). See the full lifecycle.
White Space: The data hall floor where IT equipment lives — as opposed to "gray space," which houses the electrical and mechanical plant that supports it.
Rack Density: Power draw per rack, the variable that sizes everything upstream. Traditional enterprise racks run 8–17 kW; AI training racks now demand 80–150 kW+ and force liquid cooling.
DCIM: Data Center Infrastructure Management — the software layer monitoring power, cooling, and capacity across the facility, and the operator's tool for finding stranded capacity.
MOP / SOP / EOP: Method, Standard, and Emergency Operating Procedures — scripted, rehearsed steps for every action taken in a live facility. They exist because most outages are caused by human error, not equipment failure.
Load Bank: Banks of resistive heaters that simulate IT load, letting power and cooling systems be tested at full capacity during commissioning — before a single real server is at risk.

Definitions get you to the door. The course gets you inside.

First Megawatt teaches all of this interactively — 7 tracks, 38 modules, 271 lessons built by an operator with two decades in the industry. Free, no signup required.

Start the free course → Get updates by email