Facility & site

Site model

Where the primary data center and its DR target live. This choice decides who owns power, cooling, and physical security, your capex/opex mix and staffing, your carrier options, and which side of the availability gap you start on. Retail and wholesale definitions are from Uptime Intelligence. The trade-offs are editorial and carry no figures. No pricing. Sources checked 5 Oct 2026.

Options

Retail vs wholesale is a difference in what you buy and how it is priced, not a fixed size threshold. Size boundaries quoted in kW come from brokers and marketing pages, so this page uses Uptime’s description instead: racks or cages versus a dedicated suite. A small in-house team usually pushes tier-1 workloads toward colo, because colo moves generator, UPS, and cooling maintenance to the provider. The cost is contract risk (renewal pricing, exit cost) and dependence on remote hands. Small, unstaffed edge sites (micro-DC, prefabricated or on-demand capacity) are a separate model; see Services.

Where onsite and colo diverge

The first decision splits the rest of the build. This table covers what changes; each row links to the page that owns the detail.

DR location

Facility decides where the DR target is and what rating it needs. RPO, RTO, replication, and runbooks are on Backup, and the links between sites are on Network: Edge.

• Shared risk. Check that the two sites do not share a utility grid segment, flood plain, seismic zone, or the same carrier routes. Distance alone does not prove independence.
• Synchronous replication. Find out whether any workload needs synchronous replication. If one does, the replication product’s latency limit caps how far apart the sites can be. The requirement is on Backup and the DCI design on Network: Edge.
• Rating relative to the primary. If DR is how you cover a facility outage at the primary, the DR site has to be up when the primary is down. A DR site with the same single points of failure as the primary gives less cover than its distance suggests.
• Cloud DR. You inherit the provider’s regional design and service SLAs. You cannot inspect or certify its facilities, so judge it on published architecture and on your own failover tests.

Site selection

Use this checklist for your own building and for checking a colo’s location. ANSI/BICSI 002-2024 (published May 2024) is the most detailed public reference for site selection and space planning. It covers design methodology, site selection, structural and architectural requirements, and the building systems. It is paywalled.

• Utility capacity and lead time. Ask the utility what capacity is available and when. Uptime’s 2026 outage analysis notes “worsening grid constraints” (Uptime press release, May 2026).
• Natural hazards. Flood, seismic, and severe-weather exposure (seismic design is under Seismic)
• Nearby hazards. Neighbouring industrial sites, flight paths, rail and pipelines. Take separation distances from BICSI 002 or your insurer, not from rules of thumb.
• Fibre and carriers. Which carriers already have fibre nearby, and over how many diverse routes (Network: Edge)
• Access and staffing. Road access for deliveries and fuel, and travel time for your staff and maintenance contractors

Site model design / build / manage

Design

• Decide onsite, colo, or hybrid against your staffing, capex appetite, and the resilience target in Resilience
• Pick the DR location with shared-risk checks, then hand the RPO/RTO work to Backup
• For colo, decide retail or wholesale by the terms you need, not by a kW threshold

Build

• Onsite: confirm utility capacity and carrier laterals in writing before committing to the building
• Colo: run the colo evaluation before signing

Manage

• Revisit the site model at each major refresh or contract renewal
• Re-check DR shared-risk assumptions when either site, its utility, or its carriers change

Resilience & facility rating

How much redundancy the facility has, what it protects against, and how a rating proves it. Sources are Uptime Institute and TIA pages, a TÜV NORD summary of EN 50600, and Uptime’s outage research. The standards themselves (Uptime Tier Standard: Topology, ANSI/TIA-942-C, EN 50600) were not read in full: TIA and CENELEC texts are paywalled and the Uptime standard requires registration. Where this section relies on a secondary source it says so. Power and cooling topologies that deliver these levels are on Power and Cooling. Sources checked 5 Oct 2026.

Redundancy: components vs paths

N is the capacity needed to carry the full load. N+1 adds one independent spare component, for example five units where four carry the load. 2N duplicates each system in full (Vertiv, educational article). The vocabulary counts boxes. The ratings below also count distribution paths: the switchboards, transfer switches, UPS output bus, static bypass, cable routes, and chilled-water pipework that connect the boxes to the IT load.

A networking analogy is close: N+1 components on one path is like redundant line cards and power supplies in a single chassis. You can survive a card failure and swap a card live, but backplane or supervisor work still needs an outage. Two paths is two chassis with dual-homed hosts.

The “closest rating level” column follows the level definitions below: all three schemes put the step from level 2 to level 3 at the second distribution path, not at more spare components. A matching configuration is not a rating; certification also checks many details this table leaves out.

Concurrent maintainability vs fault tolerance

Concurrently maintainable means planned work causes no outage. Uptime: “A Tier III data center requires no shutdowns for equipment replacement and maintenance. A redundant delivery path for power and cooling is added.” Fault tolerant means an unplanned single failure causes no outage. Uptime: “Fault Tolerance means that when individual equipment failures or distribution path interruptions occur, the effects of the events are stopped short of the IT operations” (Uptime Journal, updated Apr 2021).

The practical consequence: a level-3 facility is not fault tolerant. While one path is out for maintenance, the site runs on a single path, and a failure in that window can drop load. Ask any provider how it schedules and controls that exposure.

Rating schemes compared

Three schemes are in common use. Their levels line up broadly at the path boundary, but they are analogous, not equivalent: scope and criteria differ, especially on generators, physical security, and telecom spaces. Do not translate a certificate in one scheme into a level in another.

Uptime Institute Tiers. Cover site-infrastructure topology: power and cooling. Operations are assessed separately (see certification below). Availability percentages are not part of the scheme: “In 2009, Uptime Institute removed all references to ‘expected downtime per year’… The current Tier Standard of Topology does not assign availability predictions to Tier levels” (Uptime Journal). Per-Tier availability percentages that still circulate are legacy figures, not Tier properties. Uptime’s 2018 Topology update clarified “the need for on-site power generation” (trade press, Mar 2018), and a vendor author citing an Uptime paper reports that for Tier III and IV “the generator, not the utility, is to be treated as the primary source of power” (CSE Magazine, 2014). Both are secondary sources.

TIA-942. The current revision is ANSI/TIA-942-C, published May 2024. Its changes touch “telecommunications, power, cooling, architecture, fire protection, safety, and physical security” (TIA), so a TIA-942 rating says more than an Uptime Tier about telecom spaces and the building. TIA dropped the word “Tier” for “Rated” from its 2014 edition at Uptime’s request, and uses Arabic numerals where Uptime uses Roman (EPI, a TIA-licensed certification body, 2017). The standard is paywalled.

EN 50600. The European series. Besides availability classes it defines protection classes (unauthorized access, fire, environmental events) and granularity levels for energy-efficiency monitoring (TÜV NORD). Its international counterpart is the ISO/IEC 22237 series, which TÜV USA describes as “the direct evolution and international equivalent of the European EN 50600 standard” (TÜV NORD / TÜV USA, a certifier and so a secondary source). There is no single scheme owner: certification is offered by bodies such as TÜV, so check the certifier, the parts and classes in scope, and the date.

Certified vs “designed to”

A rating means something only when an independent body has awarded it for a named facility and the award is current. Both Uptime and TIA certify the design and the constructed facility separately, and both let awards lapse.

TIA-942 audits must be done by a TIA-licensed certification body, not by individual auditors, and TIA publishes a registry of certified data centers (TIA). In 2015 Uptime stopped issuing design-only certifications to North American commercial data center providers, citing their use as marketing for facilities that had not been certified as built (Data Center Knowledge, Jul 2015).

Wording on this site. “Tier III” or “Rated-3” is used only for a facility holding a current certificate. An uncertified design is described by its intent: “concurrently maintainable design intent” or “fault-tolerant design intent”. “Tier III-class” or “Rated-3-class” is used only as a contract target, meaning equivalent intent that the provider must back with evidence. It does not mean certified. A provider’s “built to Tier III standards” without a current constructed-facility certificate is a claim, not evidence.

Decision point: the onsite availability gap

Operations matter as much as topology

Ratings describe the plant, not how it is run. Uptime: “the long-term value of any site is determined by how well it performs in production, which is largely a function of how it is operated” (Uptime Journal). Its 2026 analysis found that “Failures to follow established procedures remain the leading driver of human error-related outages” (Uptime press release, May 2026), and the 2025 edition found human error involved in two-thirds to three-quarters of outages (Data Center Knowledge, May 2025). Concurrent maintainability only helps if maintenance follows written procedures that switch paths correctly. Runbooks, change control, and monitoring are on Ops.

Resilience design / build / manage

Design

• State the target as a level and a scheme (for example “Rated-3-class, concurrently maintainable design intent”), not as a percentage
• Onsite: mark every shared path element on the one-line diagram and choose one of the three gap options in writing
• Decide early whether you will certify. Certification has to be planned from design documents onward.

Build

• Test every path and transfer under load before go-live. The power and cooling systems are on Power and Cooling.
• If you claim a rating, obtain the constructed-facility certificate; a design certificate alone does not prove the build

Manage

• Keep the single-point-of-failure register current after every change. A TCCF lapses when capacity components or paths change.
• Track TIA surveillance audits and certificate expiry dates
• Review outages and near misses against the stated design intent

Building spaces

The rooms a data center needs, their purpose, and where they sit. Facility owns the rooms, their sizing, location, and separation. What goes inside is owned elsewhere: pathways, cabinets, and the structured cabling plant on Racks; carrier ordering, demarcs, and the cross-connect process on Network: Edge; access control on Security. Functional-area definitions are from a vendor summary of TIA-942-A, because the standard is paywalled. Sources checked 5 Oct 2026.

White space and gray space

White space is the computer room where IT equipment runs. Gray space is the support space: electrical rooms, UPS and battery rooms, mechanical plant, the generator yard, the loading dock, and staging and storage. “Gray space” is industry usage rather than a term defined in TIA-942. Plan both together: gray space grows with IT load, and adjacency matters for cable and pipe runs, equipment delivery routes (doors, corridors, lifts), and keeping maintenance staff out of the computer room. Plant sizing is on Power and Cooling.

Entrance room

The entrance room is “where the service enters the data center. It contains demarcation points to the service provider’s network and backbone cabling to other buildings” (Fluke Networks summary of TIA-942-A, 2018). As a space, decide how many entrance rooms you need, where they sit relative to the main distribution area, and who may enter them, since carrier technicians need access. Put the building-entry points for each entrance room (conduit sleeves and wall or floor penetrations) on separate building faces, so the pathways do not converge before they reach the rooms. Diverse entrances and routes are the core of carrier diversity; route diversity, separation figures, and the audit are on Network: Edge.

Meet-me room

In a colo, the meet-me room (MMR) is the provider-run room where carriers and customers interconnect. It is the colo’s shared entrance room. As a space, ask how many MMRs the building has, whether they are separated (different floors, fire zones, and entrances), which building entrances feed each MMR, and who controls access. Ordering and routing cross-connects through them is on Network: Edge.

Distribution areas

TIA-942 names the spaces that structure the cabling hierarchy. Here they matter as rooms or zones to plan for; the cabling inside them is on Racks.

Building spaces design / build / manage

Design

• Size white space and gray space together, for the growth you plan, not just day 1
• Plan entrance rooms and their separation together with the carrier diversity design on Network: Edge
• Place each entrance room’s building-entry points (conduit sleeves, penetrations) on a separate building face so pathways do not converge before the rooms (Network: Edge)
• Check delivery routes (doors, corridors, lifts, ramps) for the largest and heaviest items you will install

Build

• Confirm entrance-room access rules with carriers before they install
• In colo, get the MMR count and separation in writing

Manage

• Keep room drawings and their access lists current
• Keep gray space clear; do not let it turn into storage that blocks maintenance

Structure & seismic

Whether the building can carry the equipment and keep it in place. This section is a set of questions for a structural engineer, not design values: the governing standards are paywalled and the values depend on jurisdiction. Rack weight ratings and cabinet seismic kits are on Racks; airflow under a raised floor is on Cooling. Sources checked 5 Oct 2026.

Floor loading

Loaded cabinets, UPS battery cabinets, and liquid-cooling units concentrate weight on small footprints. Check both the uniform load (spread over the floor area) and the concentrated load (point loads under feet and casters), and the delivery route as well as the final position. TIA-942 sets a floor-loading criterion for Rated-3. A TIA-licensed certification body says auditors may accept a lower floor capacity if the operator shows it has a process to keep the floor from being overloaded (EPI FAQ, a secondary source; the standard is paywalled). Have a structural engineer confirm the actual slab. In colo, ask for the hall’s rated floor load in writing.

Raised floor vs slab

As a building decision, a raised floor needs extra slab-to-slab height, ramps or level changes at doors, and seismic bracing for its pedestals, and the floor tiles and pedestals become part of the load path. A slab (hard floor) puts equipment directly on the structure and moves cable and pipe routes overhead. Whether underfloor air supply still makes sense is a cooling decision (Cooling); pathways are on Racks.

Seismic

In the US, building codes reference ASCE 7, whose Chapter 13 covers nonstructural components such as equipment and its anchorage. ASCE 7-22 changed how seismic forces on equipment anchorage are calculated (Simpson Strong-Tie engineering blog, May 2026, a secondary source). In Europe the counterpart is Eurocode 8 (EN 1998). Questions for the structural engineer and the building official:

• Which code edition applies, and what seismic design category the site falls in
• Whether the owner or the jurisdiction will treat the data center as essential, which raises the design forces on equipment and anchorage
• Whether active equipment (UPS, generators, cooling units) needs seismic certification from its manufacturer
• How cabinets, raised-floor pedestals, overhead pathways, and battery racks are anchored or braced

Structure design / build / manage

Design

• Give the structural engineer worst-case cabinet, battery, and plant weights with footprints, including growth
• Decide raised floor vs slab together with cooling and pathway design
• Confirm seismic code edition and design category early; they affect anchorage and equipment choice

Build

• Inspect anchorage and bracing before handover
• Colo: get the hall floor rating in writing and plan heavy cabinets within it

Manage

• Re-check floor loading when density rises or liquid cooling is added
• Keep the as-built anchorage record with the facility documents

Colo evaluation

How to check a colocation provider’s rating claim, read its SLA, and build the contract checklist. Providers are not ranked here. The SLA example is one provider’s public SLA, used to show what to read, not as a benchmark. Contracted kW, density, fit-out, cross-connect process, and remote access design are owned by other pages and linked. No pricing. Sources checked 5 Oct 2026.

Retail vs wholesale terms

Retail colo is bought by the rack or cage at a fixed fee based on maximum power capacity and space, under a standardized contract. Wholesale is bought as a dedicated suite on negotiated terms, which can include a fixed capacity payment, metered IT power passed through, cooling fees calculated as actual power times an agreed design PUE, utilization factors, blended fixed and variable models, or build-to-suit and powered-shell arrangements (Uptime Intelligence, Dec 2025). Model every option against your expected power use, not just the headline rate.

Evidence for a claimed rating

Rank what the provider shows you, strongest first:

On multi-building campuses one hall can be certified and another not, so match the certificate to the space you will occupy. Compliance attestations such as SOC 2 reports and ISO/IEC 27001 certificates cover the provider’s controls (security and availability processes), not its power and cooling topology. You will still want them as evidence for your own audits. Physical security controls are on Security.

Reading the SLA

A colo SLA usually measures something narrower than the rating suggests. One large provider’s published colocation SLA is a useful public example:

Credits are small next to outage cost. In Uptime’s 2025 surveys, 57% of respondents said their most recent major outage cost more than $100,000, and about one in five said more than $1 million (Uptime press release, May 2026). Treat an SLA as a signal of the provider’s confidence, not as insurance.

Contract checklist

• Rating evidence for your hall, per the evidence ladder, with a duty to tell you if a certificate lapses
• Contracted power: kW per cabinet or cage, A/B feeds, and whether it is breaker-based or metered (Power)
• Density and containment rules (Cooling)
• Cage fit-out rules (Racks)
• Carrier neutrality: how many carriers and cloud on-ramps are present, and whether you may bring any carrier
• Cross-connects: price model, lead time, a written diversity commitment, and the fibre type offered for cross connects, since some providers are single-mode only (Network: Edge)
• Remote hands: scope, hours, response commitment, and billing, and whether no-fault-found tests are billable (Network: OOB)
• Access and escort rules, access lists, and shipping, receiving, and storage (Security)
• Maintenance notices: notice period, and how path switching is communicated
• Incident reporting: a committed delivery time for root-cause reports
• Compliance reports available to you for your own audits
• Expansion rights or right of first refusal on adjacent space
• Term, renewal price escalation, early termination, and exit (decommissioning and removal obligations)

Colo evaluation design / build / manage

Design

• Put rating evidence, SLA terms, and the contract checklist into the RFP
• Read the SLA definitions and exclusions before comparing headline percentages
• Plan dual-corded equipment on A/B feeds so the SLA actually applies

Build

• Check the certificate against the issuer’s records and walk the hall you will occupy
• Confirm A/B feed provisioning, contracted kW, and cross-connect diversity at turn-up

Manage

• Track maintenance notices, incident reports, and SLA credits against the contract
• Watch certificate validity and surveillance audits
• Start the renewal review well before the notice date