How to choose a data centre
Picking a colocation or data centre provider is a multi-year, hard-to-reverse decision — and the cheapest quote is rarely the right one. This is a vendor-neutral guide to what actually matters: resilience, power, connectivity, compliance, sustainability and commercials — and the questions that separate a sound facility from a costly mistake.
Why the Choice Compounds
A data centre decision isn’t a procurement line item — it sets the ceiling on your power, density, connectivity and compliance for years, and it’s expensive to undo. Three things make getting it right worth the effort up front.
Downtime is costly
Even a well-run Tier III facility permits planned and unplanned interruption. An outage in a poorly-chosen site — single power feed, weak SLA — can cost far more in a single incident than the premium for a better facility ever would.
Exit is a project
Leaving a data centre means re-cabling, cross-connect migration, hardware moves and downtime risk. Choose as if you’ll be there 5+ years, because switching is rarely quick or cheap.
It shapes what’s next
The power headroom, rack density and connectivity you accept today constrain what you can deploy tomorrow — including AI/GPU workloads that need far more power and cooling than standard racks.
Score Every Facility the Same Way
Shortlist objectively: score each facility against the same weighted criteria rather than reacting to the slickest tour or the lowest headline price. Below is the weighting we use as a starting point — tune it to your workload (a latency-sensitive trading platform weights connectivity higher; an AI build weights power and cooling higher). Need candidate sites to score? Explore real facilities on our Data Centre Map →
Selection scorecard — relative weighting
Sustainability is weighted “rising” deliberately — it’s climbing fast up buyers’ lists as reporting obligations (and energy costs) grow.
Tiers & Redundancy
“Tier” is the most quoted — and most misused — word in the industry. The Uptime Institute’s classification describes a facility’s capability, not a marketing badge. What matters is whether the site can be maintained, and survive a failure, without taking your kit down.
| Tier | What it actually means | Redundancy | Commonly-cited availability |
|---|---|---|---|
| Tier I | Basic capacity — single path, no redundancy | N | ~99.671% |
| Tier II | Redundant capacity components | N+1 (partial) | ~99.741% |
| Tier III | Concurrently maintainable — maintain any component without downtime | N+1, dual paths | ~99.982% |
| Tier IV | Fault tolerant — survives any single failure, no interruption | 2N / 2N+1 | ~99.995% |
The availability percentages are commonly cited industry references, not official Uptime Institute figures — the Tier standard defines capability (concurrently maintainable vs fault tolerant), not an uptime number. For most enterprise workloads, Tier III is the sweet spot; truly mission-critical systems justify Tier IV. Beware “Tier III-equivalent” claims with no certification behind them.
N — no margin
Exactly enough capacity to run the load. Any component failure or maintenance means downtime. Fine for dev/test; not for production.
N+1 — one spare
One redundant unit per system (e.g. an extra UPS or CRAC). Lets the facility maintain or lose one component without dropping the load. The practical baseline.
2N — fully duplicated
Two complete, independent systems (A+B power, cooling, paths). Survives a full system failure. The standard for fault-tolerant, Tier IV-class sites.
Power, Cooling & Efficiency
This is where facilities genuinely differ — and where AI workloads are rewriting the requirements. Ask hard questions about capacity, redundancy and how efficiency is measured.
Power capacity & density
Standard racks draw ~5–10kW; AI/GPU racks can demand 40–130kW. Confirm the per-rack limit, total contracted power, and whether there’s headroom to grow — not just today’s footprint.
Power redundancy
Look for diverse utility feeds (A+B), UPS with the right autonomy, generators with a credible fuel contract, and evidence of real maintenance and load testing — not just kit on a sheet.
Cooling
Hot/cold-aisle containment as a minimum; free-cooling for efficiency; and — increasingly essential — a clear path to liquid cooling if you’ll run high-density or AI hardware.
Efficiency (PUE)
A useful comparator, but ask how it’s measured (annualised, at realistic load). Lower PUE means more of your power bill goes to compute, not overhead.
PUE benchmark — where a facility sits
Connectivity, Location & Compliance
A resilient facility you can’t reach (over the network or in person), or that fails your auditors, is the wrong facility. Three areas to pin down.
Connectivity
Carrier-neutral with multiple carriers and diverse, redundant fibre paths. Check cloud on-ramps (AWS Direct Connect, Azure ExpressRoute), available cross-connects, and what they cost — recurring cross-connect fees add up.
Location & jurisdiction
Latency to your users and staff; natural-hazard and flood risk; power cost and grid stability; and data-residency / jurisdiction for your compliance regime.
Security & compliance
Physical: 24/7 staffed, mantraps, biometric access, CCTV, zoned floors. Certifications to insist on: ISO 27001 (security), SOC 2, PCI DSS where relevant, ISO 50001 (energy), ISO 22301 (continuity). Ask to see current certificates, not logos on a website.
SLA & support
An uptime guarantee with meaningful penalties (not token credits), defined remote-hands availability and response times, and a clear escalation path. Read what the SLA actually commits to.
Commercials & Red Flags
The headline rate hides most of the real cost. Understand the model, the escalators and the exit before you sign.
Pricing model
Per-kW, per-rack or per-sq-ft — and is power passed through at cost or marked up? Factor in cross-connect fees, remote-hands rates and setup. Compare like for like across your shortlist.
Contract & exit
Term length, annual price escalators, notice periods, and — critically — de-install and exit terms. A great rate with a punishing exit clause isn’t a great deal.
Red flags worth walking away from
Vague SLA, token penalties
An uptime “promise” with no real financial consequence is marketing, not a guarantee.
Single feed or no generator
One utility feed, or no credible standby power and fuel plan — a single point of failure for your whole estate.
No carrier neutrality
Locked to one network provider means captive pricing and a connectivity single point of failure.
Opaque power billing
If they won’t explain exactly how power is metered and charged, assume it’s not in your favour.
No room to grow
No power or space headroom means you’ll be migrating again sooner than you think.
Won’t show evidence
Reluctance to share current certifications, audit reports, or allow a proper site visit is the biggest tell of all.
How Optronix Helps
Choosing well is a lot of technical due diligence — much of it on-site, and easy to get wrong when you’re comparing a landlord’s sales deck rather than the facility itself. Selection is the first stage of the data centre lifecycle, and where we start with most clients: we help you shortlist, audit and choose data centres across the UK, Europe and worldwide — then stay with you through build and live operations.
Requirements & feasibility
We translate your workload into real numbers — power, density, space, connectivity and compliance needs — so you shortlist against the right criteria.
Independent facility audits
We assess shortlisted sites against the criteria in this guide — resilience, power, cooling, security — objectively, as your advisor, not the facility’s.
Selection & negotiation
Score, compare and pressure-test commercials and SLAs so the decision is evidence-based and defensible.
Design & delivery
Once chosen, we design the white space and run the build, cabling and migration — the rest of the lifecycle, under one roof.
Managed services & live operations
Once you’re live, we keep it running — managed services, DCOps, smart hands and managed field services, with ongoing support across our global coverage. One partner for the whole lifecycle, not a roster of vendors.
Prefer we run the comparison for you?
As an AVANT partner we use Pathfinder to search 1,600+ data centres and evaluate 330+ providers on real data and live pricing, across connectivity, colocation, cloud and more. Data-backed recommendations built around your needs.
Choosing or auditing a data centre?
Bring us your shortlist or your requirements, and we’ll run an independent assessment against the criteria in this guide — no facility has a stake in our answer except yours.
Sources & notes
- Uptime Institute Global Data Center Survey 2024 & 2025 — industry-average PUE of 1.56 (flat for five consecutive years)
- Uptime Institute Tier Classification System — definitions of Tier I–IV (capability-based)
- Tier availability percentages (~99.671%–99.995%) are commonly-cited industry references (e.g. TechTarget, Colocation America), not official Uptime Institute figures
- Rack power-density ranges and certification standards per general industry practice (ISO 27001, SOC 2, PCI DSS, ISO 50001, ISO 22301)
Independent guidance. This guide is vendor-neutral and recommends no specific facility or provider. Weightings are a starting framework to adapt to your own workload and risk profile, not a scoring standard.