A data centre company offers a place for digital services to stay available, protected and manageable. Customers rarely see the power rooms, cooling systems, network connections, maintenance schedules or access logs. Yet those systems decide whether a payment service, school platform or business application keeps working when something changes.
Consider Meridian Compute House, a fictional Singapore colocation operator serving several unrelated clients. During a facilities maintenance event, both customer power feeds briefly become unavailable because an overlooked common dependency affects the same critical path. The data centre reports that its backup equipment responded, but several customers still experienced service interruptions. The example is invented; it does not describe an actual facility or incident.
An earlier Gold Standard of Data Centres guide explains facility capacity and continuity. This article, within the Gold Standard master hub, concerns the company: what it promises, what it operates, the risks it retains when contractors intervene, and how it balances reliable computing infrastructure with energy efficiency, physical security and customer obligations.
Find the part of data centre management you need
Follow a critical power dependency from the contract through a service interruption and verified operational improvement.
Chapter index
Commercial promise, site and shared dependencies · Chapters 1–5
Security, maintenance and live infrastructure · Chapters 6–10
Continuity, measurement and sustainability · Chapters 11–15
People, contracts and future resilience · Chapters 16–19
1. A data centre company sells usable infrastructure, not racks
Colocation, wholesale data centres, cloud platforms, managed hosting providers and enterprise-owned server rooms can look similar but sell different products. A colocation landlord may supply space, power, cooling and cross-connects without administering the customer’s servers. A cloud provider controls a different layer of software and services. The customer needs clear boundaries.
Meridian’s contract states which equipment the company owns, who controls the racks, who is responsible for servers and networks, and how maintenance changes will be authorised. When a customer’s application becomes unavailable, support must distinguish facility events from client configuration without shirking evidence gathering. A company earns trust by making its genuine responsibility clear before a failure occurs.
Back to chapter index · Next chapter
2. Check the site, power and operating approvals before sales
A data centre site relies on dependable utility capacity, cooling provisions, fire-safety arrangements, network access, lawful premises use and investment permissions appropriate to the jurisdiction. Having floor space does not automatically mean the company can power or cool the proposed IT load.
Singapore’s IMDA Green Data Centre Roadmap describes how the country is managing capacity growth and efficiency. Meridian’s hypothetical expansion team should distinguish a site proposal from approved and usable electrical capacity, and verify applicable planning and safety requirements with competent specialists. Sales should never promise a powered rack before the critical dependencies can be delivered.
Back to chapter index · Next chapter
3. Map every critical dependency before claiming redundancy
Resilience is more than installing two items with different labels. Separate power feeds or network paths can still share an upstream transformer, common cable corridor, maintenance system or control point. A company claiming redundancy must define which failure cases the architecture actually tolerates and which it does not.
Meridian’s incident reveals that two feeds depended on a component believed to be independent. The operations team investigates drawings, operating records and tested scenarios instead of describing the event only as ‘unexpected downtime’. The company should qualify the redundancy it sells and communicate known shared dependencies to customers at a level appropriate to security and commercial confidentiality.
Back to chapter index · Next chapter
4. Specify contracted capacity and the conditions behind it
Customer demand can be measured in rack space, electrical power, IT equipment consumption, network connectivity and other service quantities. These measures should not be conflated. A facility’s total connected capacity is not the same as the capacity that can be safely supplied to every customer at once.
Meridian verifies the agreed load per customer area and how expansions are assessed, with qualified engineering input. A client adding more compute equipment can change heat, airflow and electrical demand. The provider should document what is permitted, what triggers technical review, and which claims depend on future upgrades. Capacity accuracy protects the customer’s planning and the facility’s safe operation.
Back to chapter index · Next chapter
5. Manage cooling as an availability and efficiency responsibility
IT equipment produces heat, and inadequate cooling can reduce reliability or availability. Cooling systems also consume substantial electricity and, in some designs, water. Operators need to manage temperature and humidity within conditions suitable for their real installed equipment rather than copy a universal setpoint from another facility.
Singapore’s Tropical Data Centre Standard SS 697:2023 provides a methodology for safely considering higher operating temperatures in tropical conditions. The guidance describes potential cooling energy savings but does not authorise a specific site’s settings without appropriate evaluation. Meridian tests changes under qualified oversight and monitors customer equipment outcomes before claiming an efficiency improvement.
Back to chapter index · Next chapter
6. Make physical security an accountable chain of custody
Data centres can hold sensitive commercial equipment, media and network infrastructure. Entry controls should distinguish authorised employees, client representatives, vendors and visitors. Controls need to work at night and during emergencies, including when normal access systems are unavailable.
Meridian documents access authorisation, escorted work, relevant logs and customer notification where required by contracts. A contractor badge may give legitimate entry to a shared service area but not to a customer’s locked cage. The company should periodically review permissions and handle departures promptly. Physical security that exists only on the installation day is not a dependable protection system.
Back to chapter index · Next chapter
7. Define who may perform maintenance and when
Data centres require planned inspection and replacement of electrical, mechanical, fire-safety and IT-support equipment. Some work carries serious hazards and must follow the qualified professional and legal process for the particular facility. Commercial pressure to avoid an outage cannot substitute for the proper authority.
Meridian’s maintenance calendar shows the proposed scope, approved method, dependent customer services, qualified workers, communications and a means of stopping the task when unexpected conditions appear. A completed work order should describe what was verified after the intervention. This is management guidance, not a procedure for switching electrical power or changing live infrastructure.
Back to chapter index · Next chapter
8. Test recovery against the exact service customers need
Backup generators, UPS equipment, diverse networks and contingency procedures exist to protect continuity. Testing that an individual piece of equipment starts is not the same as proving the customer’s application remained supported through the intended failure scenario. A company needs risk-appropriate integrated evidence.
The eduKate case study about a generator that started while essential circuits stayed dark illustrates this problem. Meridian’s leadership should ensure test plans reflect actual customer dependencies, with qualified execution, credible records and a process to correct gaps. Passing a single component check cannot prove uninterrupted business service.
Back to chapter index · Next chapter
9. Treat the network as a customer-facing system
An available server without adequate connectivity may be useless to its owner. Network performance depends on carriers, cross-connects, routing, equipment and the customer’s configurations. Data centre operators need to define which network services they sell and which belong to third-party providers.
Meridian’s support desk maintains a fault map for the parts it controls and coordinates with external carriers where appropriate. A link marked ‘up’ does not necessarily guarantee the customer’s traffic reached its destination. Customers should receive a verified explanation and an escalation path rather than repeated claims that the building’s infrastructure looks normal.
Back to chapter index · Next chapter
10. Manage remote hands with identity and change controls
Colocation customers may ask local technicians to inspect equipment or perform defined actions when they cannot attend the facility. Such work can create risk if the instruction arrives from the wrong person or is misunderstood. Authorisation, task scope and evidence matter.
Meridian verifies the customer’s designated contacts and logs each approved request. A remote-hands technician should not interpret a vague message as permission to alter another customer’s system. After authorised work, the provider records what was done and the observable outcome, escalating conditions outside the agreed scope. Trustworthy assistance is about precision and permissions, not just fast response.
Back to chapter index · Next chapter
11. Measure availability with a visible denominator
Availability numbers can refer to power at a socket, cooling service, a cross-connect or an entire application. A company must say what the percentage measures, which period it covers, and which events are included or excluded. One aggregate figure may hide a small group of customers with repeated disruption.
Meridian offers a fictional 30-day report: a customer’s contracted facility service was unavailable for 90 minutes out of 43,200 minutes. Simple availability is (43,200 − 90) ÷ 43,200, approximately 99.79%. That is an illustrative calculation, not a statutory data centre standard or a determination of contractual credits. The company should separately investigate timing, cause, customer impact and remedies.
Back to chapter index · Next chapter
12. Understand PUE before promising green performance
Power Usage Effectiveness compares total data-centre energy use with IT equipment energy use over a defined measurement scope and period. A lower ratio generally indicates lower facility overhead relative to IT consumption, but it does not by itself establish total carbon emissions, responsible water use, renewable electricity sourcing or a particular customer’s computing efficiency.
Illustration: a facility uses 14,000 kWh overall while its IT equipment uses 10,000 kWh, yielding PUE 1.40. If the same IT work and consumption were supported with total energy of 13,000 kWh under comparable conditions, PUE would be 1.30 and the arithmetic difference would be 1,000 kWh. This is not a forecast or a guarantee that an individual change would save that energy.
Back to chapter index · Next chapter
13. Use current Singapore standards without overstating certification
Singapore’s IMDA Green DC Roadmap identifies BCA-IMDA Green Mark for Data Centres 2024, the Tropical DC Standard SS 697:2023 and SS 715:2025 for more energy-efficient IT equipment. They describe different parts of efficiency and assurance rather than one universal award.
The 2025 DC-CFA2 factsheet sets particular energy and Green Mark requirements for applicants to that capacity-allocation exercise. Meridian should not describe those tender-specific thresholds as a universal legal requirement for every existing data centre. Any certification claim must identify the actual facility, standard, tier and evidence.
Back to chapter index · Next chapter
14. Report energy, water and carbon in separate measures
Cooling upgrades, IT refreshes, power contracts and water systems can affect different environmental outcomes. A company could lower its PUE while still growing absolute electricity use by selling more IT capacity. Water and energy trade-offs must be considered in the actual design and climate context.
Meridian reports total electricity use, IT energy, relevant water information and supported environmental claims with clear boundaries. Purchasing an environmental certificate is not the same as installing a physical renewable generator on the site. Management should describe energy procurement accurately and avoid vague claims that every customer workload is carbon neutral merely because one sustainability programme exists.
Back to chapter index · Next chapter
15. Protect customer information without claiming to control all data
Colocation operators may not have access to customer application content but can still process identity, access, incident and billing records. Managed hosting providers may control additional data. Contracts and operational procedures should reflect the actual responsibility and law applicable to each layer.
Meridian restricts facility records to necessary roles and secures remote access to its own management systems. It must coordinate appropriate incident notifications without giving one tenant unnecessary information about another. The Cybersecurity Companies guide addresses the separate discipline of security services. A data centre company must still protect its own access and operational technology.
Back to chapter index · Next chapter
16. Keep staff and contractors competent across every shift
Facilities operation is continuous, but specialist support may not be on site around the clock. The company needs shift handovers, training, competent on-call arrangements, fatigue management and a clear route to stop risky work. A facility without a qualified person available to decide an urgent issue is more fragile than its equipment diagrams imply.
Meridian’s night team should know which alarms require immediate specialist escalation and which customer communications must be sent. Managers review repeated false alarms rather than training everyone to ignore them. Contractor responsibilities remain visible in incident reports; outsourcing maintenance must not make organisational learning impossible.
Back to chapter index · Next chapter
17. Explain service-level agreements without hiding essential exclusions
Availability, response times, maintenance notices, service credits and liability limits differ between contracts. A customer may believe a quoted uptime applies to all dependencies while the company promises only an individual service component. A gold-standard company makes the scope, exceptions and remedy process understandable before the purchase.
Meridian explains exactly which service the SLA covers, how an incident can be reported and how disputed measurements will be assessed. It does not promise an application-level guarantee over software it never operates. When a credit is due under agreed terms, the billing system should be able to apply it correctly and show how it was calculated.
Back to chapter index · Next chapter
18. Plan capacity growth without undermining current customers
AI computing and cloud services can increase demand for rack space, power, cooling and networks. A company must balance expansion revenue with the consequences of construction, energy supply and maintenance for existing tenants. A signed contract for future capacity is not proof that the facility can safely supply it today.
Meridian’s board reviews resource constraints, upgrade schedules, resilience risks and financing before approving another hall. The Singapore Green Data Centre Roadmap gives national context for energy-aware expansion. Responsible growth preserves current service obligations while testing whether new capacity can satisfy regulatory and operational conditions.
Back to chapter index · Next chapter
19. The gold standard survives a common-mode failure
Meridian’s fictional incident showed that two labelled power feeds were not operationally independent under the tested circumstance. It reviews the shared dependency, revises maintenance authorisation, improves customer communication and tests the relevant corrective measures under qualified supervision. A claim of ‘backup power activated’ is not sufficient when tenants still lost service.
Data centre companies earn trust through clear ownership of service layers, reliable physical security, trained workers, measured availability, controlled change and honest energy reporting. The real gold standard is a facility the customer can use and recover from, not a collection of impressive numbers displayed in a tour.
The energy calculation a data centre customer should understand
These are invented numbers under stated assumptions. An actual PUE calculation needs properly defined meter boundaries and measurement periods.
| Metric | Illustrative calculation | What it does not prove |
|---|---|---|
| IT equipment energy | 10,000 kWh | Not total building consumption |
| Total facility energy, scenario A | 14,000 kWh | PUE = 1.40 |
| Total facility energy, scenario B | 13,000 kWh | PUE = 1.30 |
| Difference at same assumed IT work | 1,000 kWh | Not a site-wide annual savings guarantee |
| Scope beyond PUE | Water use, carbon and customer uptime | Require separate evidence |
Facility and commercial audit
- Can the operator identify exactly which service layer it owns?
- Have shared power, cooling and network dependencies been reviewed rather than just labelled redundant?
- Are maintenance and remote-hands actions performed by authorised competent people?
- Can backup systems demonstrate the intended customer outcome under meaningful tests?
- Does the SLA identify the actual availability denominator and exclusions?
- Are PUE and absolute energy consumption both measured and described accurately?
- Are tropical efficiency standards and certifications cited with their correct scope?
- Do physical access, client separation and data handling survive a real incident?
- Is future capacity sold only after technically and commercially credible assessment?
Singapore standards and related Gold Standard articles
- IMDA: Green Data Centre Roadmap — national efficiency framework
- IMDA: Tropical Data Centre Standard SS 697:2023 — operating-temperature methodology
- IMDA: Green Data Centre Standard — continuous improvement
- IMDA: DC-CFA2 2025 requirements — particular capacity-allocation criteria
- eduKate: Data Centres — capacity and continuous service
- eduKate: Backup generator case study — tested essential load
- eduKate: Cybersecurity Companies — authorised incident response
This is an educational company-management guide, not a data centre design, electrical operating instruction, safety certification or contract opinion. Qualified personnel must determine actual facility settings, maintenance and redundancy tests. The company and figures are fictional.