Clectiv
Reference

Colocation glossary for buyers

Definitions written for the person reading their first colocation quote. Where a term is routinely misused in provider marketing — and several are, systematically — the entry says what the term actually means and what the marketing usually means instead.

Terms#

Grouped loosely by what they describe — space, power, interconnection, resilience, compliance and commercial terms — but each definition stands alone and can be read in any order.

Colocation#
Also called: colo
Renting space, power and cooling in someone else's data center for hardware you own and operate. You buy the building, the electricity and the physical environment; the servers, the operating systems and the applications remain yours.
Distinct from cloud (you rent the compute, not the space) and from managed hosting (someone else operates the hardware). The three are frequently mixed in one estate.
Cabinet#
Also called: rack, cab
The smallest standard unit of colocation: a lockable enclosure, conventionally 42 to 52 rack units tall, that you fill with your own equipment. Pricing is usually quoted per cabinet per month plus power, and partial cabinets are sometimes sold by the rack unit.
Rack unit#
Also called: RU, U
The vertical measurement unit of a cabinet: one rack unit is 1.75 inches. A server described as "2U" occupies two of them. Relevant to buyers mainly when leasing partial cabinets, which some operators price per rack unit.
Cage#
Private floor space enclosed by mesh partitioning inside a shared data hall, holding multiple cabinets. Bought when physical separation is required for security or compliance reasons. The buildout is a project with its own cost, quoted separately from the recurring charge.
Suite#
Also called: private suite, data hall
A fully enclosed private room within a facility, with its own walls and access control. The step above a cage in isolation and cost, and the point at which deployments start being quoted like small wholesale deals.
Retail colocation#
Colocation sold by the cabinet, cage or suite, typically on 12- to 36-month terms, with the operator supplying the fitted-out space. The product most buyers below roughly a megawatt are shopping for.
The boundary with wholesale is set by each operator, not by an industry standard — operators place it anywhere from roughly 1 MW to 10 MW. If your requirement sits near a given operator's line, ask for it quoted both ways.
Wholesale colocation#
Colocation sold in megawatt blocks or as dedicated data halls, on long terms — commonly five to fifteen years. Priced per megawatt of capacity rather than per cabinet, and generally cheaper per kW than retail because the operator supplies less.
Powered shell#
A building delivered with power and structure but without the data-center fit-out, which the tenant completes. The lowest-cost, highest-effort end of the market, used by buyers with their own build capability.
Build-to-suit#
Also called: BTS
A facility or data hall developed to a single anchor tenant's specification. Timelines run from roughly six months to over two years, and the commercial structure is a negotiated development agreement rather than a standard colocation order.
Raised floor#
The elevated flooring in a data hall through which cooling air and cabling historically ran. As a spec figure it matters because it usually describes the actual technical space, which can be a small fraction of a building's gross square footage — a building measured in hundreds of thousands of square feet may hold far less usable data-hall area.
Campus#
Multiple buildings operated by one company on one site. A recurring source of confusion in facility counts: an operator may count a campus as one location while an independent count treats each building separately, so two accurate counts of the same estate can differ substantially.
kW and MW#
Kilowatts and megawatts — the electrical capacity of a deployment, and the primary commercial unit of colocation. A deployment is defined by its committed power far more usefully than by its cabinet count, because two deployments with the same number of cabinets and different densities are different products.
Density#
Also called: kW per rack, kW per cabinet
How much power a single cabinet draws. Standard enterprise density has historically sat in the low single-digit kW per cabinet; modern high-density and AI deployments run to tens of kW and beyond. Density decides which buildings can host you, whether liquid cooling is required, and what you pay per kW.
Committed power#
Power capacity you contract and pay for whether or not you draw it. The dominant retail model. Over-committing on nameplate ratings rather than measured draw is one of the most common and most expensive colocation mistakes.
Metered power#
Power billed on actual measured consumption rather than on reserved capacity. Frequently sold as a hybrid: a committed baseline with metered billing above it. Ask which model a quote uses — the same headline rate produces very different bills under each.
Breakered power#
Power sold as circuits of a stated amperage rather than as measured consumption, with usable draw conventionally capped below the breaker rating for safety headroom. Where it applies, you are buying the circuit, so unused capacity is not refunded.
N+1#
Redundancy with one spare unit beyond the minimum needed to carry the load — four cooling units where three would do. Protects against a single component failing, and does not protect against a whole distribution path failing.
2N#
Full duplication: two complete independent systems, each able to carry the entire load. More resilient and more expensive than N+1. Buying 2N for an application that tolerates N+1 is a common overspend, and worth deciding deliberately rather than by default.
You will also see 2N+1 and 2N+2 used to describe electrical and cooling topologies with duplication plus additional spare capacity.
PUE#
Also called: Power Usage Effectiveness
Total facility energy divided by the energy delivered to IT equipment. A PUE of 1.5 means half again as much power is consumed by cooling and losses as reaches your servers. Lower is better, 1.0 is the unreachable floor.
Relevant to you only if you pay for power based on facility consumption, or if efficiency is a procurement requirement. Compare like with like — annualised figures and design targets are different claims, and a design target is not a measurement.
WUE#
Also called: Water Usage Effectiveness
Water consumed per unit of IT energy delivered. Increasingly material in water-stressed metros, where it can also carry regulatory and reputational weight independent of cost.
Liquid cooling#
Cooling that carries heat away in liquid rather than air — direct-to-chip or immersion. Effectively mandatory above the densities modern AI hardware reaches. Whether a specific building supports it is a facility-by-facility question, not a portfolio claim.
Cross-connect#
Also called: XC
A physical cable — fibre or copper — between your equipment and another party's inside the same facility: a carrier, a cloud on-ramp, or another tenant. Billed monthly per connection, plus an installation charge.
The most commonly under-modelled cost in colocation. In dense deployments cross-connect spend can rival cabinet spend, you frequently pay at both ends of a connection, and there is often a fee to remove one before term end. Count the connections you will actually run and price them over the full term.
Meet-me room#
Also called: MMR
The neutral room in a facility where carriers and tenants physically interconnect. Its density — how many networks terminate there — is much of what distinguishes an interconnection-rich building from ordinary data-center space.
Carrier hotel#
A building whose primary value is the accumulated concentration of networks in it rather than its data-center product. Typically decades old, often in a downtown core, and effectively impossible to replicate: the fibre routes converge there because they always have.
Carrier-neutral#
Also called: network-neutral
The operator does not own or mandate the networks you use, and permits any carrier to be brought in. The opposite is a facility where connectivity must be bought from the operator. Most commercial colocation is carrier-neutral, but confirm which carriers are actually on-net in the specific building.
On-net#
A carrier is on-net in a building when its network already terminates there and you can reach it with a cross-connect. A carrier not on-net must build in, which costs money and months. Ask for the on-net list for your building, not for the operator's portfolio.
Cloud on-ramp#
A private connection into a cloud provider's network — AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect, Oracle FastConnect — bypassing the public internet for lower and more predictable latency.
COMMONLY MISUSED. "On-ramp available" usually means reachable from the operator's site over their backbone or through an interconnection partner, not physically present in your building. Both work and they differ in latency, hop count and who you call when it breaks. Ask specifically whether the on-ramp for your cloud region is in this building or reached over a network.
Internet exchange#
Also called: IX, IXP
Shared infrastructure where many networks peer with each other directly, usually settlement-free, instead of paying transit. Presence of a major exchange in a building is a genuine and checkable ecosystem advantage for traffic-heavy deployments.
Point of presence#
Also called: PoP
A location where a network operator has equipment — often a rack inside somebody else's data center. A PoP is not colocation you can buy.
Worth knowing because it systematically distorts published facility counts: third-party directories that ingest listings by address count network nodes as data centers, which can inflate an operator's apparent footprint by a factor of two or more. If a listed "facility" cannot be bought as colocation, it is not supply.
Software-defined interconnection#
Also called: NaaS, virtual cross-connect
Provisioning connectivity through a portal rather than by ordering a physical cable — an operator's own fabric, or a third-party network-as-a-service provider. Priced separately from physical cross-connects, usually per port and bandwidth tier. Faster to turn up; an additional hop and an additional counterparty.
Dark fibre#
Unlit fibre leased between locations, with the buyer supplying the equipment at both ends. Used to tether buildings on a campus into one logical deployment, and to reach a carrier hotel from a cheaper building nearby.
Uptime Institute Tier#
Also called: Tier I, II, III, IV
The recognised third-party data-center resilience classification, awarded by the Uptime Institute per facility, in ascending order of redundancy and concurrent maintainability. A certified facility has a certificate and a number.
THE MOST MISUSED TERM IN THIS GLOSSARY. "Tier III design," "designed to Tier III standards," "Tier III equivalent" and "Tier III quality" are all self-declared and involve no certifying body. "Tier 5" does not exist as an Uptime classification — where you see it, it is a proprietary standard. Genuine certification is far rarer than the market implies, including among the largest operators. Ask for the certificate number and the facility it names.
Concurrently maintainable#
The property that any single component can be taken out of service for maintenance without interrupting the load — the practical distinction that matters most day to day, because planned maintenance is far more frequent than component failure.
SOC 2#
An attestation, by an independent auditor, that an organisation's controls over security and related criteria were operating as described. A Type 2 report covers a period of operation rather than a point in time, and is the more meaningful of the two.
Ask for the report or the auditor letter, and check what it scopes: attestations are frequently limited to named facilities and can cover an estate an operator no longer owns. A logo on a trust page is not an audit letter.
ISO 27001#
An international certification of an information security management system. Like SOC 2, it is scoped — confirm the certificate covers the facility you are buying, not just the corporate entity.
PCI DSS#
The payment card industry's data security standard. A facility can supply the physical and environmental controls a cardholder-data environment depends on, but the facility being compliant does not make your deployment compliant — the scope split is yours to confirm.
HIPAA and BAA#
The US health information privacy regime. There is no such thing as a HIPAA certification. What matters commercially is whether the operator will sign a Business Associate Agreement, and what it covers — so ask for the BAA rather than for a HIPAA claim.
FedRAMP and ATO#
The US federal authorisation programme for cloud services, and the Authority to Operate issued under it, at Moderate or High baselines.
Routinely misread in colocation. An authorisation may cover an operator's software products and not its data centers; where it does cover facilities it is usually scoped to specific named buildings rather than the portfolio. Ask which artefact is authorised, at what baseline, and for which facility.
MRC and NRC#
Monthly recurring charge and non-recurring charge — the ongoing rent and the one-time costs. Non-recurring charges cover installation, cage or suite buildout, power provisioning and cross-connect installation, and are quoted separately from the recurring charge if you ask for them separately.
Escalator#
A contractual annual increase in the recurring charge, standard across the colocation market and essentially never published. It compounds over the term and is negotiable. Get the percentage in writing before signing, and model total contract value with it applied.
Evergreen auto-renewal#
A contract that renews automatically unless cancelled within a defined notice window. Close to universal in retail colocation. Establish the notice period at signature, put a reminder in ahead of it, and ask whether renewal pricing is capped.
Remote hands#
Basic physical tasks performed by facility staff on your equipment — reboots, cable checks, visual inspection. Billed hourly with a minimum, in defined increments, at a higher rate outside business hours. Prepaid blocks are commonly available at a discount.
Smart hands#
Technical on-site work beyond remote hands — installation, configuration, troubleshooting. Priced above remote hands. Where an operator draws the line between the two is a commercial detail worth confirming, because it decides which rate applies to routine work.
SLA and service credits#
The committed service level and what the operator pays when it misses. The credit schedule is the part that matters: a headline uptime commitment with a trivial credit schedule transfers essentially no risk. Ask for the schedule in writing, and ask what the process is to claim against it.
Total contract value#
Also called: TCV
The full cost over the committed term including escalators, all non-recurring charges and realistic interconnection spend. The only figure that meaningfully compares two differently-structured colocation offers, and the one providers are least likely to lead with.