Colocation

Overview#

Colocation lets a customer place its IT equipment in a third-party data centre and pay for the surrounding infrastructure as a service. The operator provides the physical envelope: power feeds, cooling, fire suppression, physical security, networking access, and the building itself. The customer brings servers, storage, networking, and the operational expertise to run them.

It is the dominant deployment model for any organisation that needs the resilience and bandwidth of a real data centre but does not have the scale, capital, or appetite to build one. UK enterprise IT, financial services, regulated workloads that cannot live on public cloud, and a large fraction of AI inference build-outs all rely on colo.

Retail vs Wholesale#

What's Included (and What Isn't)#

• Included: physical space, power delivery to rack PDU, cooling, fire suppression, perimeter security, building maintenance, common-area utilities.
• Sometimes included: cross-connect cabling, smart-hands, basic remote-hands, environmental monitoring.
• Not included: customer IT, internet transit, customer cabling within rack/cage, customer-side compliance evidence beyond facility-level certifications.
• Optional add-ons: bandwidth (transit / IX), DDoS mitigation, structured cabling design, project management, escorted access for customer staff.

AI's Impact on Colo Economics#

Colo facilities were largely designed for 5-15 kW per rack. AI workloads land at 40-120 kW per rack, breaking the cooling assumptions and the power-density model. Operators have responded in three ways:

• Dedicated AI suites — retrofitted halls with direct-to-chip cooling, busbar power, and 400/800 G optical paths.
• Greenfield AI builds — entirely new facilities engineered for 100 kW racks from day one.
• Tiered pricing — AI-class power is priced differently from general-purpose colo power, often at a premium reflecting the cooling cost.
• Some operators decline AI workloads in legacy halls because the heat exceeds the air-handling envelope of the room.

When to Use Colo (vs Cloud, vs On-Prem)#

• vs public cloud: colo is cheaper at scale (above ~6-12 months continuous full utilisation), gives full hardware control, and supports regulatory constraints cloud cannot.
• vs on-prem: colo eliminates building, utility, generator, and chiller capex; saves on staffing; provides better connectivity and physical security than most enterprise rooms.
• vs hyperscale-build: colo is the right answer when you need 1-50 racks; you build at hyperscale when you need 10+ MW continuous.
• AI-specific: colo is often the fastest route to 100 kW-rack capacity in a chosen jurisdiction, because the operator has already done the utility, planning, and cooling work.

Operational Pitfalls#

• Power contracting: 'up to X kW' is not the same as 'reserved X kW'. Burstable contracts can be capped at peak times. Read the small print.
• Cross-connect pricing: in many retail colos, recurring cross-connect fees rival the rack rental over a 3-year term. Model the full cost.
• Service levels: colo SLAs cover the facility, not the customer's IT. SLA credits for facility outages can be modest relative to the business impact.
• Access control: physical access procedures vary materially between operators. Build them into runbooks.
• Density growth: a rack contracted at 10 kW that grows to 30 kW may require relocation or upgrade — and colos may charge to reassign space.
• Exit costs: migrating out of a colo at end-of-term has real cost — equipment movement, cross-connect re-establishment, and potential dual-running.