Watt & Rack
Sheet Colocation, explainedFacilities in dataset 6,018Operators 2,387
Home / Colocation

What colocation is, and how to buy it

Colocation is renting a place for your own servers inside someone else's data center. You are not buying computing. You are buying three physical things, power, space and connectivity, and the contract is mostly about how much of each you get and how reliably.

Guide · buying colocation

Colocation in one paragraph

A colocation provider builds and runs the facility: the utility connection, transformers, uninterruptible power supplies and generators, the cooling plant, fire suppression, the security perimeter and the meet-me room where carriers and clouds connect. You rent a slice of that infrastructure, bring your own hardware, and pay monthly. The provider keeps the building alive; you keep your servers alive. Everything else, from who reboots a frozen machine at 3 a.m. to who pays when a power feed fails, is a matter of contract.

What you actually rent

Power

Power is the expensive part and the part that scales. It is delivered to your rack as one or more circuits, usually in redundant pairs called A and B feeds, and sold either as an allocation you may draw (kW, or amps at a stated voltage) or as metered energy. When you read a price, find the power first. On the published cards we track, US cabinet prices work out at USD 212 to 472 per kW of allocation, while the size of the cabinet barely matters. The pricing sheet shows the full set.

Space

Space comes in steps: a few rack units (U) in a shared cabinet, a quarter, half or full cabinet, a caged area holding several cabinets, or a private suite or hall. One U is 1.75 inches of vertical rack space, and a standard full cabinet holds around 42 to 48U. Space is cheap relative to power in most markets, but it is not free everywhere. In Singapore, for instance, a published half rack costs more per kW than a quarter rack because the extra money buys space rather than power.

Connectivity

Your rack has to reach the outside world. Small offers bundle a port, a traffic allowance and a few IP addresses. Larger deployments buy IP transit from carriers in the building and pay the provider for cross-connects, the physical cables from your rack to a carrier, an internet exchange or a cloud on-ramp. Carrier choice in the building, and the price of each cross-connect, often matter more over a contract than the rack price itself.

Cooling, security and the building

Cooling capacity limits how much power a rack can actually use, so it is sold implicitly with power. Physical security, access control, fire suppression and staffing come with the building. These are what certifications and audits describe, and what an SLA promises.

Retail versus wholesale

Retail colocation is sold by the U, rack or cage. Power, space and often some bandwidth are bundled into a monthly price, contracts can be month to month or one to three years, and the provider may offer remote hands, meaning on-site staff who rack, cable or reboot equipment for you.

Wholesale colocation leases a block of committed IT load, typically hundreds of kilowatts to megawatts, often as a dedicated suite or hall. It is quoted per kW per month, energy is metered and billed on top at or near cost, and terms run longer. Across the 22 markets CBRE prices in our data, wholesale asking rents sit between USD 160 and USD 403 per kW per month at the midpoint. The jump from retail to wholesale is where the per-kW price falls the most, which is why a growing deployment should ask when it qualifies.

Cabinets, cages and suites

Shared rack (per U). Your server sits in a cabinet with other customers' equipment. Cheapest entry, little control over physical access, power measured in watts or a fraction of an amp.

Quarter, half and full cabinets. A lockable cabinet, or a lockable section of one, with its own power circuits. This is the default unit of retail colocation and the unit most rate cards price.

Cage. A fenced area with several cabinets and its own access control. Cages let you place your own cabinets, run your own cabling between them and meet audit requirements that call for physical separation.

Private suite or hall. A walled room, sometimes with dedicated cooling and power infrastructure. This is where wholesale begins, and where the contract starts to look more like a property lease than a service agreement.

Power density: the question that decides everything

Density is the power drawn per rack, in kW. It decides which facilities can take you, how many racks you need and which format you buy. A cabinet of ordinary servers and network gear may draw a few kW; GPU servers draw many times that. Of the 90 facilities in our dataset that publish a maximum rack density, the median is 15 kW. Above roughly what a building's air cooling can remove, a provider needs rear-door heat exchangers or direct liquid cooling; 304 facilities in our data list liquid cooling.

Two numbers to get right before you ask for prices: the total IT load you will actually draw, and the density per rack. The rack count follows from those two. Asking for "ten racks" without a kW figure invites quotes that cannot be compared.

Also ask what the power figure means. In North America a circuit's continuous load is limited to 80% of the breaker rating under the electrical code, so a 30 A, 208 V circuit gives about 4.99 kW of usable power, not 6.24. Rate cards that state kW have usually done that arithmetic; quotes in amps may not have.

Service levels and certifications

An SLA defines what the provider promises and what you get when it fails: typically power availability per feed, environmental ranges for temperature and humidity, and response times for remote hands. Read three things in any SLA: how availability is measured (per feed or for the rack with both feeds), what the credit is (usually a share of the monthly fee, not your losses), and what you must do to qualify, such as using both A and B feeds correctly.

Certifications describe the building and the operator's controls. Uptime Institute's Tier system rates the design and construction of the facility's infrastructure: Tier III means maintenance can be done without shutting down the IT load, Tier IV that the infrastructure tolerates a single fault. Tier ratings describe topology, not a guaranteed uptime percentage. SOC 2 Type II and ISO 27001 cover security controls and processes; PCI DSS matters if you handle card data. In our dataset, 1,110 facilities mention an Uptime tier or a Tier III/IV level, 684 SOC 2, 978 ISO 27001 and 781 PCI DSS, out of 6,018. Only 618 show an Uptime award for the constructed building or its operations; the rest rest on design awards, other standards or the operator's own words. Ask for the current report or certificate, not the logo on a website: our guide to data center tiers shows how to check one.

Contract terms that move the price

  • Term. Longer terms buy lower monthly prices or waived setup fees, but published cards rarely state a discount; the term is negotiated.
  • Power commitment. All-in allocation, committed capacity with metered energy, or a mix. Know which one you are signing.
  • Escalation. Annual price increases, often a fixed percentage or tied to inflation, and how energy price changes pass through.
  • Setup and installation. One-time fees for the cabinet, power circuits and each cross-connect.
  • Expansion rights. A right of first refusal on adjacent space or extra power, so growth does not mean a second site.
  • Remote hands. Included hours, hourly rate after that, response time.
  • Exit. Notice period, renewal terms (auto-renewal is common) and what happens to your equipment at the end.

Colocation, cloud or your own room

Cloud rents you computing by the hour and hides the hardware; it wins for spiky, uncertain or short-lived workloads. Colocation rents you the building and leaves the hardware to you; it wins when the load is steady, when you own the equipment already, or when hardware like GPUs is cheaper to own than to rent. An on-premises server room wins only when you need equipment physically next to people or processes, because few offices can match a data center's power redundancy, cooling and carrier choice. Many buyers end up with a mix: steady workloads in colocation, cloud for the rest, connected by a cross-connect to a cloud on-ramp in the same building.

How to buy colocation, step by step

Start with the load, not the location. Measure or estimate the kW you will draw, the density per rack and how much it will grow over the term. Then choose the markets that work for latency, staff access and carriers. Collect quotes from at least three providers on the same written requirement, so the numbers compare. Compare them per kW, with cross-connects, setup and escalation included, over the full term. Visit or audit the shortlisted sites. Negotiate the term against the setup fees and expansion rights, then sign.

Colocation buying checklist

  1. Total IT load in kW, today and at the end of the term.
  2. Density per rack in kW, and whether any racks need liquid cooling.
  3. Power model: all-in allocation or committed capacity plus metered energy, and the rate per kWh if metered.
  4. Redundancy: A+B feeds, and how much load each feed must carry alone.
  5. Space format: per U, cabinet, cage or suite, and room to expand next to it.
  6. Carriers and clouds in the building, and the monthly and one-time price per cross-connect.
  7. Certifications you need, with the current report or certificate.
  8. SLA: what is measured, the credit and the conditions.
  9. Remote hands: included hours, hourly rate, response time.
  10. Term, annual escalation, renewal and notice period.
  11. Setup fees for the cabinet, circuits and cross-connects, and what is waived for a longer term.
  12. Access rules for your staff and shipping and receiving for equipment.

Colocation questions

What is colocation in simple terms?

Colocation means renting space, power and cooling for servers you own inside someone else's data center. You bring and manage the hardware; the provider runs the building, the power chain, cooling, physical security and the meet-me room where networks connect.

What is the difference between colocation and cloud?

In the cloud you rent computing as a service and never touch the hardware. In colocation you own or lease the servers and rent only the place to run them. Colocation suits steady, predictable workloads, owned hardware such as GPU servers, and cases where you want control of the physical machines.

What is the difference between retail and wholesale colocation?

Retail colocation is sold by the U, rack or cage with power, space and often bandwidth bundled into a monthly price. Wholesale colocation leases larger blocks of committed IT load, quoted per kW per month, with energy metered on top and longer contracts.

What is managed colocation?

Managed colocation adds operational services from the provider on top of the space and power, such as monitoring, hardware replacement or operating-system management. One published card in our data prices its managed per-kW tier at EUR 50 above the unmanaged rate.

Want to see where facilities are before you shortlist markets? The atlas at megawattlas.com maps them.

Sources Uptime Institute, NFPA 70, ISO/IEC 27001 and SOC 2

Send one requirement, get comparable quotes

Describe the load and the markets once. We take it to providers that fit and bring back quotes priced the same way, so you can compare them per kW.

kW
racks

Your request comes to the Watt & Rack team, not to a mailing list. We match the requirement with providers that fit it and reply by email. Need more detail? Use the full request form.See privacy.