45 terms

Data center glossary

Plain-language definitions with real numbers. Free to cite with attribution.

2N redundancy

2N redundancy means a data center installs two complete, independent systems, each capable of carrying the full load alone — effectively 100% duplication of UPS plants, generators, distribution paths, or cooling.

Bare metal

Bare metal is a dedicated physical server rented to a single customer with no virtualization layer imposed by the provider — the customer gets direct access to the hardware and full control of the operating system.

Behind-the-meter power

Behind-the-meter (BTM) power is electricity generated at or adjacent to the consumption site — on the customer's side of the utility meter — so it flows to the data center without passing through the public grid.

Build-to-suit (BTS)

Build-to-suit (BTS) is a development model in which a data center developer constructs a facility to a specific tenant's requirements — location, capacity, density, cooling technology, and security specs — under a long-term lease signed before or during construction.

Carrier-neutral data center

A carrier-neutral data center is a facility whose owner does not favor or restrict access to any single telecommunications carrier, allowing tenants to choose freely among all network providers present in the building.

Colocation

Colocation is a service where a data center operator rents space, power, and cooling to customers who install and operate their own servers and network equipment.

CRAC / CRAH units

CRAC (computer room air conditioner) and CRAH (computer room air handler) units are the workhorses of data center air cooling.

Cross-connect

A cross-connect is a physical cable link — fiber or copper — run between two customers or between a customer and a carrier inside the same data center, typically routed through the meet-me room.

Dark fiber

Dark fiber is installed but unused (unlit) optical fiber that a customer leases and lights with its own transmission equipment, rather than buying a managed bandwidth service.

DCIM (Data Center Infrastructure Management)

DCIM (data center infrastructure management) is software that monitors, measures, and manages a data center's physical infrastructure — power, cooling, space, and assets — bridging the gap between facilities and IT systems.

Direct-to-chip cooling

Direct-to-chip (DTC) cooling circulates liquid coolant through cold plates mounted directly on the hottest components — CPUs, GPUs, and memory — removing 70-80% of server heat at the source while fans handle the remainder.

Edge data center

An edge data center is a smaller facility located close to end users or devices — in secondary cities, at network aggregation points, or at cell-tower sites — to minimize latency and backhaul costs.

GPU cloud

A GPU cloud is a cloud service that rents GPU compute capacity — for AI training, fine-tuning, inference, and rendering — billed per GPU-hour or via reserved contracts.

Greenfield development

Greenfield development is the construction of a data center on previously undeveloped land, as opposed to brownfield conversion of existing industrial buildings or expansion of operating campuses.

HBM (High Bandwidth Memory)

HBM (high bandwidth memory) is a type of DRAM in which memory dies are stacked vertically and connected to the processor over a very wide interface through a silicon interposer, delivering far higher bandwidth than conventional DDR or GDDR memory.

Hot/cold aisle containment

Hot/cold aisle containment is a data center layout practice that physically separates cold supply air from hot exhaust air to prevent mixing and recirculation.

Hyperscale data center

A hyperscale data center is an extremely large facility built by or for major cloud and internet platforms — AWS, Microsoft, Google, Meta, Alibaba, and peers — designed to scale to tens or hundreds of megawatts of IT load across standardized, software-managed infrastructure.

Immersion cooling

Immersion cooling submerges entire servers in a non-conductive dielectric fluid that absorbs heat directly from all components.

Interconnection

Interconnection is the direct exchange of traffic between networks, clouds, and enterprises inside or between data centers, bypassing the public internet for better performance, security, and cost.

Internet exchange (IX / IXP)

An internet exchange (IX or IXP) is a shared switching fabric where multiple networks — ISPs, content providers, CDNs, and enterprises — interconnect to exchange traffic directly, a practice called peering.

IT load

IT load is the electrical power actually consumed by computing equipment — servers, storage, and network gear — measured in kW or MW, excluding cooling, power-conversion losses, and building systems.

kW per rack (rack density)

kW per rack, or rack density, is the amount of electrical power a data center delivers to a single cabinet, which in turn determines how much compute can be installed there.

Latency

Latency is the time it takes data to travel between two points on a network, usually measured in milliseconds as round-trip time (RTT).

Liquid cooling

Liquid cooling is the use of fluids instead of air to remove heat from IT equipment, exploiting the fact that water conducts heat roughly 25 times better than air and carries far more heat per unit volume.

Meet-me room (MMR)

A meet-me room (MMR) is a secure, dedicated space inside a data center where telecommunications carriers, ISPs, cloud providers, and tenants physically interconnect their networks via cross-connects.

Micro data center

A micro data center is a self-contained, factory-integrated unit — often a single enclosure or container — that packages racks, UPS, cooling, fire suppression, security, and monitoring into a footprint of roughly 1-10 racks and 5-100 kW.

N+1 redundancy

N+1 redundancy is a design approach in which a system has one more component than the minimum needed to carry the load: if N units are required, N+1 are installed.

Neocloud

A neocloud is a new generation of cloud provider focused almost exclusively on GPU compute for AI, in contrast to hyperscalers' broad service catalogs.

NVLink

NVLink is NVIDIA's proprietary high-speed interconnect that links GPUs to each other (and to NVIDIA CPUs) at far higher bandwidth than PCIe, letting multiple GPUs pool memory and act as one large accelerator.

PDU (Power Distribution Unit)

A PDU (power distribution unit) is equipment that distributes electrical power to IT hardware in a data center.

Powered shell

A powered shell is a data center building delivered with the core structure, secured utility power connection, and basic site infrastructure in place, but without the tenant-side fit-out — no UPS, generators, cooling plant, or racks.

PPA (Power Purchase Agreement)

A PPA (power purchase agreement) is a long-term contract — typically 10-20 years — under which a buyer commits to purchase electricity from a specific generation project at agreed prices.

Prelease

A prelease is a lease commitment signed for data center capacity before the facility is completed — sometimes before construction begins.

PUE (Power Usage Effectiveness)

PUE (Power Usage Effectiveness) is a ratio that measures data center energy efficiency by dividing total facility power by the power delivered to IT equipment.

Rack unit (U)

A rack unit (U or RU) is the standard vertical measurement for equipment mounted in a data center rack, equal to 1.75 inches (44.45 mm).

Raised floor

A raised floor is a data center flooring system of removable tiles (typically 60 x 60 cm) supported on pedestals 30-120 cm above the structural slab, creating a plenum used to distribute cold air through perforated tiles and to route power and network cabling.