Colocation Pricing in 2026: What AI Companies Actually Pay

Colocation pricing is opaque by design. Providers rarely publish rates, quotes vary wildly depending on who is asking, and the true cost of a deployment is scattered across line items that are easy to overlook. For AI companies evaluating GPU colocation, understanding the real cost structure is essential to making an informed infrastructure decision.

This guide cuts through the ambiguity. We share real pricing benchmarks gathered from dozens of provider conversations across the UK market, explain every cost component, flag the hidden fees that inflate your monthly bill, and show you how to negotiate a better contract.

How Colocation Pricing Works

Colocation is priced on two fundamental axes: space and power. Everything else is built on top of these.

Per-Kilowatt Pricing

The dominant pricing model for AI and high-density colocation is per-kilowatt per month. You commit to a specific power draw -- say, 40kW for a fully loaded GPU rack -- and pay a monthly rate per kW. This rate typically bundles the facility's space, power delivery, cooling, and basic physical security into a single figure. Per-kW pricing is the clearest way to compare providers because it normalises for density.

Per-Rack Pricing

Traditional colocation is often priced per rack per month, with a specified power allocation included. A standard 42U rack might come with 5-10kW of power for a fixed monthly fee. This model works for traditional IT workloads but becomes problematic for AI deployments because you will quickly exceed the bundled power allocation and face expensive overage charges. Always convert per-rack pricing to a per-kW figure to compare accurately.

What the Monthly Fee Includes

A standard colocation fee typically covers: physical rack space, power delivery to the rack (at your committed level), cooling to maintain operating temperatures, basic physical security (badge access, CCTV), and a baseline level of uptime SLA (typically 99.99%). What it usually does not include: network cross-connects, bandwidth, remote hands labour, setup fees, and power overages.

Standard vs High-Density Pricing

AI workloads require fundamentally different infrastructure than traditional IT, and the pricing reflects this. Here is how standard and high-density colocation compare in the UK market:

A counter-intuitive finding: the per-kW cost for high-density colocation is often lower than for standard colocation. Providers can serve more compute per square metre of floor space, which improves their economics. The absolute monthly cost per rack is higher because you are consuming 3-5x more power, but you are getting proportionally more compute capacity. This is good news for AI companies -- density actually works in your favour on a unit-cost basis.

What Drives Colocation Costs

Power

Power is the single largest cost driver, typically representing 40-60% of your monthly bill. UK electricity costs for data centres vary by region and contract, but most providers pass through energy at a markup or bundle it into the per-kW rate. In 2026, wholesale electricity prices in the UK have stabilised somewhat, but the grid connection capacity in popular data centre corridors (particularly Slough and west London) remains constrained, which keeps pricing elevated in those areas.

Cooling

Liquid cooling infrastructure carries a higher upfront cost for providers, which is reflected in pricing for high-density racks. Facilities that have invested in direct liquid cooling (DLC) infrastructure charge a premium, but this is offset by better energy efficiency. The Power Usage Effectiveness (PUE) of a well-run liquid-cooled facility can be 1.1-1.2, compared to 1.4-1.6 for air-cooled facilities, which means you waste less money on cooling energy.

Location

London and the Thames Valley corridor command the highest prices in the UK due to demand, connectivity density, and land costs. Pricing in Manchester, Edinburgh, and other regional hubs is typically 15-30% lower for comparable specifications. However, the network connectivity ecosystem in London is significantly richer, which matters if you need low-latency access to financial markets, cloud on-ramps, or internet exchanges.

Contract Length

Longer contracts deliver lower monthly rates. A 36-month commitment typically secures a 10-20% discount compared to a 12-month term. Providers prefer longer contracts because they provide revenue certainty and justify their capital investment in high-density infrastructure. For AI companies with predictable long-term compute needs, locking in a multi-year rate protects against price increases.

UK Pricing Benchmarks for 2026

Based on real quotes and conversations with providers across the UK market, here are the ranges AI companies should expect. All figures are monthly and exclusive of VAT:

Hidden Costs That Inflate Your Bill

The quoted per-kW or per-rack rate is rarely the full picture. These are the costs that surprise teams who have not negotiated colocation contracts before:

Cross-Connect Fees

Every physical connection to a network carrier, cloud provider on-ramp, or internet exchange requires a cross-connect. These typically cost £150-£300 per month per connection. A serious GPU deployment might need 3-6 cross-connects for redundancy and multi-carrier diversity. That is an extra £450-£1,800 per month that does not appear in the headline rack rate.

Remote Hands Charges

When your hardware needs physical intervention -- rebooting a hung server, swapping a failed drive, re-seating a cable -- the data centre's staff performs "remote hands" work. Most providers include a small allowance (perhaps 2-4 hours per month), but anything beyond that is charged at £50-£150 per incident or per hour. For GPU clusters with many servers, these charges can accumulate.

Power Overage Fees

If your actual power draw exceeds your contracted allocation, you will face overage charges that are significantly more expensive per kW than your base rate. Some providers charge 1.5-2x the contracted rate for overages. GPU workloads can be bursty -- a fully loaded training run draws more power than idle servers -- so right-sizing your power commitment is critical. Build in 10-15% headroom above your expected peak.

Setup and Installation

The one-time cost to prepare your cage or rack space, install power and cooling connections, and configure the physical infrastructure. For high-density GPU deployments requiring liquid cooling plumbing, this can run £5,000-£10,000 or more. Some providers waive setup fees for long-term contracts, so always ask.

Escalation Clauses

Some contracts include annual price escalation clauses tied to inflation indices (RPI, CPI) or fixed percentage increases. A 3-5% annual escalation on a 3-year contract means your year-three price is 9-15% higher than year one. Negotiate a cap or a fixed-rate guarantee for the contract term.

How to Reduce Your Colocation Costs

• Commit to longer terms: A 36-month contract will almost always secure better rates than 12 months. If your compute demand is predictable, the savings are significant.
• Negotiate power metering: Some providers offer metered power (pay for actual consumption) rather than committed power. If your utilisation varies, metered billing can save 10-20% versus over-provisioned committed power.
• Bundle cross-connects: If you need multiple cross-connects, negotiate a bundled price. Providers have margin in cross-connect pricing and will discount for volume.
• Ask about seasonal rates: Some UK providers offer lower rates for power during off-peak periods when wholesale electricity prices are lower.
• Use a broker: A specialist broker like ColoGPU has visibility across the market and can negotiate on your behalf. Providers pay the broker's commission, so there is no cost to you. Brokers also know which providers are currently offering promotional pricing or have excess capacity to fill.
• Right-size your power: Over-provisioning power wastes money. Under-provisioning triggers expensive overages. Work with your hardware vendor to model actual power consumption under typical and peak workloads, and contract accordingly.
• Consider regional facilities: If your workload does not require London's connectivity ecosystem, regional facilities in Manchester, Cardiff, or Edinburgh can offer 15-30% savings on comparable specifications.

Understanding the full cost structure before signing a contract prevents budget surprises and ensures you are comparing providers on a genuine like-for-like basis. The headline rate is just the starting point -- the total cost of ownership includes every line item discussed above.