Plain-English definitions of 43 industry terms — from PUE and Tier ratings to dark fiber and immersion cooling.
An enhanced redundancy model providing two complete N+1 redundant systems for critical infrastructure. Offers the highest level of fault tolerance, ensuring continued operations even if one entire redundant system fails. Used in the most critical datacenter environments where maximum uptime is paramount.
A redundancy configuration where every critical component has an equal backup — two complete, independent systems support the load simultaneously, either in active-active or active-standby configuration. If one entire system fails, the other carries the full load without interruption. Required for Uptime Institute Tier IV certification.
Standards published by the American Society of Heating, Refrigerating and Air-Conditioning Engineers defining acceptable temperature and humidity ranges for datacenter equipment. The A1–A4 classes cover inlet air temperatures from 15°C to 45°C; most enterprise equipment is rated A1 or A2 (max 27°C inlet). Higher ASHRAE class support allows datacenters to raise cooling setpoints and reduce energy consumption.
The routing protocol that governs how data packets are routed between autonomous systems (networks) on the Internet. Carrier-neutral datacenters typically offer direct BGP peering options, allowing tenants to announce their own IP prefixes and establish redundant internet paths. Multiple BGP peers improve resilience and reduce latency.
An enclosed metal enclosure housing IT equipment in a datacenter, also called a rack cabinet or server cabinet. Standard cabinets are 19 inches wide and 42U tall, though 47U and 48U cabinets are common. Cabinets may include integrated power distribution (PDUs), cable management, and locking doors. Power density per cabinet typically ranges from 5 kW to 30+ kW in high-density configurations.
A private, physically partitioned area within a colocation datacenter, typically enclosed by wire mesh fencing or solid walls with its own locked entrance. Cages provide tenant isolation and security beyond standard shared floor areas. Common for enterprise tenants needing multiple cabinets with strict physical access controls.
A datacenter that allows multiple competing telecommunications carriers, ISPs, and network providers to co-locate their equipment and offer connectivity services to tenants. Carrier-neutral facilities enable tenants to choose from competing providers, negotiate better pricing, and build multi-homed network connections for redundancy. Compare with carrier-specific facilities locked to a single provider.
A datacenter model where the facility provides physical space, power, cooling, and network connectivity, while tenants own and manage their own server hardware. Colocation is billed by cabinet, cage, or suite plus power consumption (kW or kVA). It sits between DIY on-premises infrastructure and fully managed cloud services, offering physical control with outsourced facility operations.
Computer Room Air Conditioning (CRAC) units use refrigerant-based compressors to actively cool air and deliver it to IT equipment. Computer Room Air Handlers (CRAH) use chilled water from a central chiller plant instead of on-board compressors, improving energy efficiency at scale. Both types are deployed in raised-floor or overhead configurations to manage airflow across datacenter rows.
A direct physical cable connection between two tenants or between a tenant and a network provider within the same datacenter, bypassing the public internet entirely. Cross-connects provide low-latency, high-bandwidth private connectivity and are commonly used to connect servers to cloud on-ramps (AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect) or to peer with other networks at an internet exchange.
A metric measuring a datacenter's carbon dioxide emissions relative to the IT equipment energy consumption. CUE is calculated as total CO₂ equivalent emissions divided by IT equipment energy (kWh). A CUE of 0 indicates carbon-neutral operations using fully renewable energy. Like PUE, lower is better, and CUE benchmarks vary significantly by geography and energy grid carbon intensity.
Installed but unused optical fiber cable that carries no active data transmission — it has no light (hence 'dark'). Datacenters and network operators lease dark fiber to organizations that want to run their own optical equipment and wavelengths over dedicated fiber paths. Dark fiber provides full control over capacity, latency, and protocols without depending on a carrier's managed network services.
Software platforms that monitor, measure, and manage datacenter physical infrastructure — including power, cooling, space, and connectivity. DCIM systems provide real-time dashboards of PUE, power consumption per rack, temperature maps, capacity utilization, and asset tracking. Advanced DCIM tools can optimize cooling automatically and model infrastructure scenarios for capacity planning.
A distributed computing model placing processing and storage closer to the physical location where data is generated — in cities, factory floors, or telecommunications infrastructure — rather than centralizing workloads in large regional datacenters. Edge datacenters are typically smaller (100 kW–5 MW), geographically distributed, and optimized for ultra-low latency applications such as autonomous vehicles, industrial IoT, and 5G network functions.
A cooling technique that uses ambient outdoor air, water, or ground temperature to cool IT equipment directly or via heat exchangers, reducing or eliminating mechanical refrigeration when outdoor conditions are favorable. Free cooling is particularly effective in northern European climates where cool ambient temperatures are available for most of the year, significantly reducing PUE and energy costs. Sometimes called an economizer or airside/waterside economization.
A diesel or natural gas generator providing backup electrical power when the utility grid supply fails. Datacenters typically deploy generators sized to carry the full critical load plus cooling infrastructure. N+1 or 2N generator configurations ensure redundancy. Generators are tested weekly under partial load and during annual full-load transfer tests. Fuel contracts guarantee minimum on-site reserves (typically 48–72 hours at full load).
A datacenter that sources all or a significant portion of its electricity from renewable energy sources such as wind, solar, or hydropower. Renewable sourcing may be achieved through direct Power Purchase Agreements (PPAs), renewable energy certificates (RECs), or location in markets with a predominantly clean grid. The renewable energy percentage on Megawatt Market listings reflects the operator's self-reported or certified renewable coverage.
A datacenter floor arrangement where server racks face alternating directions, creating dedicated cold intake aisles and hot exhaust aisles. Containment adds physical barriers (doors, ceilings, blanking panels) to prevent hot and cold air from mixing, improving cooling efficiency by 10–30%. Hot aisle containment captures hot exhaust air and returns it directly to cooling units; cold aisle containment pressurizes the intake aisle with cool air.
A datacenter facility or campus built for massive-scale operation by a single large technology company (such as AWS, Microsoft, Google, Meta, or Apple). Hyperscale datacenters are characterized by custom-designed hardware, extreme standardization, dedicated facilities from 20 MW to 500+ MW, and deep vertical integration. They are not typically available for third-party colocation — the term is sometimes misused by smaller operators to imply scale.
A physical infrastructure facility where multiple Internet Service Providers and network operators interconnect and exchange traffic directly, bypassing transit providers. Internet exchanges reduce latency, improve resilience, and lower the cost of carrying traffic between participating networks. Datacenters connected to major IXPs (Amsterdam's AMS-IX, London's LINX, Frankfurt's DE-CIX, etc.) offer tenants direct peering access to thousands of networks.
The standard unit of electrical power in datacenter contexts, equal to 1,000 watts. IT equipment power consumption and datacenter power capacity are measured in kW. A single high-density GPU server rack may consume 30–100 kW; a typical enterprise cabinet averages 5–10 kW. Datacenter pricing is often expressed as a monthly cost per kW of committed power, separate from the physical space cost.
The time delay (measured in milliseconds) for a network packet to travel between two points. In datacenter selection, latency matters for applications requiring near-real-time responses: financial trading systems require sub-millisecond latency to exchanges; gaming requires sub-20ms to players; content delivery requires sub-50ms to end users. Network latency depends on physical distance (roughly 1ms per 200km in fiber), routing hops, and congestion at interconnection points.
Cooling technologies that use liquid instead of air to remove heat from IT equipment. Direct liquid cooling (DLC) pipes chilled water directly to cold plates on CPUs and GPUs. Immersion cooling submerges servers in dielectric fluid (mineral oil or engineered fluids), enabling power densities above 100 kW per rack — 10× higher than air-cooled limits. These technologies are increasingly required for AI training clusters with dense GPU configurations.
A dedicated space within a carrier-neutral datacenter where network operators physically interconnect their equipment to exchange traffic or provide cross-connect services to tenants. The MMR houses patch panels, distribution frames, and carrier cages. Access to a well-connected MMR is a key reason tenants choose specific colocation facilities — it determines which carriers, cloud on-ramps, and internet exchanges are directly reachable.
One megawatt equals 1,000 kilowatts or 1,000,000 watts. MW is the primary unit used for large-scale datacenter transactions. Hyperscale campuses may total 200–500 MW; large wholesale datacenters offer 20–100 MW; typical retail colocation facilities range from 5–30 MW. When Megawatt Market lists a datacenter's total MW, it refers to the total critical IT load capacity of the facility.
The total installed IT power capacity available to tenant equipment in a datacenter or a specific phase of a campus, expressed in megawatts (1 MW = 1,000 kW). This is distinct from total facility power, which includes cooling, lighting, and UPS losses. MW critical load is the standard unit for wholesale datacenter transactions — a 10 MW lease typically refers to 10 MW available to servers, not total facility consumption.
A redundancy model providing one additional (backup) component beyond the minimum required to support the load. For example, N+1 cooling means if 4 CRAC units are needed to cool the floor (N=4), a 5th unit is installed and operational. If any single unit fails, the remaining 4 maintain full capacity. N+1 is the standard configuration for Uptime Institute Tier III datacenter certification.
A device that distributes electrical power from the facility's supply to individual servers and equipment within a rack cabinet. Basic PDUs are simple power strips; managed PDUs provide remote monitoring, per-outlet power measurement, and the ability to power-cycle individual ports remotely. Intelligent PDUs with outlet-level metering are essential for accurate billing in multi-tenant colocation environments.
A long-term contract (typically 10–20 years) between a datacenter operator (or large tenant) and a renewable energy generator, committing to purchase a specified volume of electricity at a fixed price. PPAs provide price stability and directly fund new renewable capacity, enabling operators to credibly claim renewable energy sourcing. PPA coverage is a key sustainability metric on Megawatt Market listings.
The most widely used datacenter efficiency metric, calculated as total facility power divided by IT equipment power. A PUE of 1.0 is theoretical perfection; modern hyperscale facilities achieve 1.1–1.2; well-operated retail colo typically achieves 1.3–1.5; older facilities may exceed 2.0. PUE below 1.3 is considered excellent; above 1.5 indicates significant cooling inefficiency. The Green Grid established PUE as the industry standard metric in 2007.
The standard unit of vertical space in a server rack or cabinet, equal to 1.75 inches (44.45 mm). A standard 42U rack provides 42 rack units of usable space. Servers are commonly 1U (thin, for dense deployments), 2U (standard servers), or 4U and above (storage, GPU systems). Colocation pricing is sometimes quoted per U for retail space, though cabinet and kW-based pricing is more common for larger deployments.
A modular flooring system suspended above the structural slab, creating a plenum space used for cooling air distribution and cable management. Cold air is delivered through perforated tiles directly beneath equipment aisles. Raised floors were the dominant cooling distribution method in traditional datacenters and remain common, though many modern high-density facilities use overhead cooling distribution instead for better airflow management.
Dual independent power supply paths (typically labeled A and B) delivering electricity to tenant equipment from separate UPS systems and generator chains. Servers with dual power supplies connect one PSU to the A feed and one to the B feed, eliminating single points of failure in the power chain. Colocation facilities that offer A+B feed capability at the rack level are essential for mission-critical deployments requiring 99.99%+ uptime.
A network architecture approach that separates the network control plane (routing decisions) from the data plane (packet forwarding), enabling centralized programmable management of network resources. In datacenter contexts, SDN enables automated provisioning of network connectivity, dynamic bandwidth allocation between tenants, and network function virtualization. Many modern datacenters offer SDN-based network services alongside physical cross-connects.
Datacenter construction and equipment mounting standards vary by seismic risk zone, which is determined by regional earthquake frequency and magnitude. High seismic zone datacenters (California, Pacific Northwest, Japan, Taiwan, New Zealand) require reinforced building structures, seismically rated floor anchoring for equipment, and emergency procedures for post-earthquake assessment. Zone compliance is typically disclosed by operators in facility documentation and affects insurance terms.
A contractual commitment between a datacenter operator and tenant specifying guaranteed service levels — typically uptime availability (e.g., 99.99% per year), response time for critical faults, and financial remedies (bill credits) when targets are missed. SLA uptime of 99.9% allows ~8.7 hours downtime annually; 99.99% allows ~52 minutes; 99.999% allows ~5 minutes. Always review remedy caps (maximum credits per incident and per year) when evaluating SLAs.
The lowest Uptime Institute tier classification, providing single-path power and cooling distribution with no redundancy. Tier I facilities support only planned maintenance activity and have no protection against unplanned outages. Expected uptime is approximately 99.671% (28.8 hours downtime per year). Tier I is rare in commercial colocation — it is more typical of small on-premises server rooms or early-stage facilities.
Uptime Institute Tier II adds redundant mechanical and electrical components (N+1) to the single distribution path of Tier I, reducing unplanned downtime. Maintenance still requires partial or full shutdown. Expected uptime is approximately 99.749% (22 hours downtime per year). Tier II is common in smaller regional colocation facilities and is suitable for non-mission-critical workloads.
The most common certification level for commercial colocation datacenters. Tier III requires multiple active power and cooling paths with N+1 redundancy, allowing any planned maintenance on any component without impacting the operational load. Unplanned failures may still cause localized outages. Expected uptime is 99.982% (approximately 1.6 hours downtime per year). Tier III is the de facto standard for enterprise and carrier-grade colocation.
The highest Uptime Institute tier classification, requiring fully fault-tolerant infrastructure with 2N redundancy on all power and cooling systems. Any single failure — including unexpected equipment fault — cannot interrupt operations. Expected uptime is 99.995% (approximately 26 minutes downtime per year). Tier IV datacenters are significantly more expensive to build and operate, and are primarily used by financial services, government, and mission-critical enterprise workloads.
An electrical device providing instantaneous backup power from batteries when the main utility supply fails, bridging the gap until generators start (typically 10–30 seconds). Datacenter UPS systems are large-scale, three-phase units rated in kVA or MW. Modern transformerless UPS designs achieve efficiencies above 96%. UPS battery capacity is usually sized to provide 5–15 minutes of full-load runtime, enough for controlled generator transfer or graceful shutdown.
Datacenter uptime is expressed as a percentage of time the facility is operational over a given period. 'Five nines' (99.999%) allows only ~5.26 minutes of downtime per year; 'four nines' (99.99%) allows ~52 minutes; 'three nines' (99.9%) allows ~8.76 hours. Note that datacenter uptime refers to facility availability (power and cooling), not application availability — your software stack introduces additional failure modes beyond the physical infrastructure.
A metric measuring the volume of water consumed by a datacenter relative to IT equipment energy consumption, expressed in liters per kilowatt-hour (L/kWh). Cooling towers, evaporative coolers, and direct evaporative cooling consume water and raise WUE. Datacenters using air-cooled chillers or dry coolers with no evaporation have near-zero WUE. Water consumption is increasingly scrutinized as datacenters expand in water-stressed regions.
Have a term we should add? Send us a suggestion, or browse the directory to put these terms into context.
We use only essential storage to keep you signed in and remember your theme. We do NOT use tracking cookies, advertising cookies, or analytics.