Behind the Connection Queue: UK Infrastructure Resilience When Power, Not Floor Space, Is the Constraint
Photo: Visitor7, CC BY-SA 3.0, via Wikimedia Commons
For most of the last decade, the UK data centre conversation was about location and latency: which campus, which region, how many milliseconds to the primary trading platform. Power was an engineering detail handled by the facilities team and reflected in the monthly invoice.
That assumption no longer holds. Connection queues in several UK regions now run to multiple years, distribution network operators are imposing curtailment conditions on new and expanded supplies, and the sites that can offer capacity today are often not the sites that offer the best latency profile. Space is rarely the binding constraint. Megawatts are.
For IT leaders, this shifts resilience planning from an exercise in geography to an exercise in energy economics. The data centre has stopped behaving like a utility and started behaving like a scarce, negotiated resource.
What constraint actually looks like on the ground
The headline figures mask a more granular reality. Capacity is not uniformly scarce across the UK; it is scarce in specific places, at specific times, under specific conditions.
Three patterns are worth understanding because they drive different planning responses.
Queued capacity. A site may hold an offer for additional power that will not be delivered for several years. Contracts signed against that assumption look firm and are not.
Conditional capacity. Some connections now carry curtailment or flexibility conditions — the operator can require load reduction at defined times, or the site must participate in a demand-side flexibility arrangement. That is workable for batch analytics and awkward for a latency-sensitive transactional system.
Stranded capacity. Older sites with existing, unconstrained supplies become disproportionately valuable, and disproportionately expensive. Renewal negotiations at these facilities have a different tenor than they did five years ago.
Add to this the practical lead times for the supporting chain — transformers, switchgear, generators — and the planning horizon for any material expansion stretches beyond a typical three-year technology roadmap.
Resilience assumptions that no longer survive
The traditional colocation resilience model assumes the facility can deliver the contracted power envelope, on demand, indefinitely, with N+1 redundancy across the electrical path. That assumption needs re-testing in three specific ways.
First, ask whether the contracted capacity is delivered or committed. The distinction sits in schedules and annexes, not in the marketing documentation.
Second, establish how redundancy behaves under a constrained grid. A facility with generator backup and full fuel autonomy behaves very differently from one that relies on a curtailment agreement to avoid exceeding its import limit. Both may claim N+1. Only one survives a multi-day regional event.
Third, test the failure modes you have not modelled. Most disaster recovery plans assume a facility-level outage with failover to a secondary site. Fewer plans address a scenario where the primary site remains operational but power-constrained, and workload has to be shed or migrated at short notice under a commercial arrangement rather than a technical one.
Placing workloads by tolerance, not by habit
Once power is the scarce input, workload placement becomes an exercise in matching consumption to tolerance. A useful three-band model:
| Band | Characteristics | Placement logic |
|---|---|---|
| Latency-critical | Transactional, sub-10ms tolerance, constant load | Retain in constrained, high-cost capacity; justify each kilowatt |
| Business-critical, deferrable | Core systems with batch windows and DR tolerance | Primary in constrained capacity, DR in a lower-cost region with genuine failover tested |
| Elastic and interruptible | Analytics, media processing, dev/test, model training | Place where power is cheap and flexible; accept curtailment and design for checkpointing |
The third band is where most organisations find headroom. Workloads that can be paused, queued or migrated on notice are ideally suited to sites with flexibility conditions attached to their supply — and often to the public cloud region that happens to have capacity. The engineering requirement is that these workloads are genuinely interruptible, which means stateless design, externalised state and tested restart procedures. Many estates claim this property and have never proved it.
A related discipline is measurement. Very few organisations can state the actual kilowatt-hour consumption of an individual application or service. Without that, placement decisions are guesswork and renewal negotiations are conducted blind. Metering at the rack and, where possible, at the platform level is now a prerequisite rather than a refinement.
Contract clauses worth arguing about
Commercial terms have not caught up with the physical reality, and the gap is where risk accumulates.
- Capacity guarantees. Insist on a schedule stating the delivered power envelope in kilowatts, the date from which it is available, and the remedy if it is not delivered on time. "Best endeavours" is not a capacity guarantee.
- Curtailment and flexibility terms. If the site participates in demand-side arrangements, the contract must specify notice periods, maximum duration, frequency limits and the commercial mechanism — credits, rebates or service credits — when load reduction is requested.
- Escalation paths. Define who is notified, in what sequence and within what window when a power event occurs. A named contact and a response-time commitment are worth more than a general commitment to keep customers informed.
- Cost pass-through. Understand precisely which energy and network charges are recoverable, how they are indexed, and whether the indexation mechanism can be changed mid-term.
- Exit and reduction rights. If capacity becomes unaffordable or the site's conditions change materially, the ability to reduce committed capacity at defined intervals is a commercial resilience feature in its own right.
Review these alongside the general exit terms already under scrutiny in most 2026 renewals. The two conversations are the same conversation.
A planning lens for the next twelve months
Four actions will put most organisations in a defensible position. Document the real power envelope across every site, distinguishing delivered from committed capacity. Meter consumption at a granularity that supports placement decisions. Re-classify workloads by interruptibility and test the claim. And renegotiate, before the renewal date, any contract whose capacity language is aspirational rather than binding.
The organisations that plan around constraint early tend to find it manageable. The ones that discover it during a failover do not.
