For most of the cloud era, the question of where to put your infrastructure was answered by software and convenience. In the UK in 2026 it is increasingly answered by electricity. Power has become the binding constraint on data centre growth - grid connections are scarce in the busiest regions, energy is a dominant share of running cost, and rack power densities are climbing fast under AI workloads. That reshapes the on-premises, colocation and cloud decision around a question few buyers used to ask: can I actually get the power, where, and at what price. Here is how UK power economics should shape where your infrastructure lives.
Power is now the constraint, not space
The old mental model of a data centre was square metres of floor. The real currency now is kilowatts. Modern servers - and especially AI-accelerated ones - draw far more power per rack than the designs most facilities were built around, so a hall runs out of power and cooling capacity long before it runs out of floor space. A rack that once drew a handful of kilowatts can now demand many times that, and the building's electrical and cooling infrastructure, not its footprint, sets the ceiling.
This flips procurement. The first question for any sizeable deployment is no longer how many U of rack space but how many kilowatts you can secure, how they are priced, and whether the cooling can remove the resulting heat. In the most in-demand UK locations, that power simply may not be available at the scale or timeline you want - which is the single biggest change in the siting decision.
The UK grid reality
Grid connection has become a genuine bottleneck in parts of the UK, particularly the established data centre clusters where demand has outrun the local network's spare capacity. New or expanded connections can face long lead times, and that scarcity feeds directly into both availability and cost. For a business planning growth, the assumption that you can simply add more power when you need it no longer holds everywhere - it has become something to secure deliberately and early.
Energy price is the other half. Electricity is a dominant and volatile component of running infrastructure, and UK prices have made power efficiency a board-level concern rather than a facilities footnote. That pushes two behaviours: consolidating onto fewer, more efficient servers to cut the kilowatts in the first place, and siting workloads where power is both available and sensibly priced. The total-cost lens that ties this together is in cloud vs on-prem TCO in 2026.
- •Secure power and grid capacity early - in busy regions it is scarce and lead times are long
- •Energy is a dominant, volatile running cost - efficiency directly cuts the bill
- •Rising rack densities mean cooling, not floor space, often sets the practical ceiling
On-prem, colo or cloud: a power-led decision
Power economics change the classic deployment trade-off. On-premises gives you control but puts the grid connection, electrical capacity and cooling squarely on you - viable where your site has the headroom and the densities are modest, painful where it does not. Colocation effectively buys access to power, cooling and grid capacity that would be slow or impossible to provision yourself, which is precisely why it is the pragmatic answer for higher-density and growth scenarios in constrained regions. Cloud abstracts power entirely into a usage charge, trading capital and operational control for the provider's ability to absorb the power problem on your behalf.
Seen through the power lens, the decision often becomes: keep modest, stable, low-density workloads on-prem where the site allows; place dense, growing or AI workloads in colocation where power and cooling are already solved; and use cloud for elastic or bursty demand where you are content to pay for someone else carrying the energy and capacity risk. It is rarely all-or-nothing - most UK organisations land on a deliberate mix, and our services team helps model it.
Cooling, density and efficiency
The same AI-driven densities that strain power also strain cooling. Beyond a certain rack power, conventional air cooling reaches its limits and liquid cooling becomes necessary - which most existing on-prem rooms are not built for, but which purpose-built colocation increasingly offers. So density is not just a power question; it is a cooling-capability question that often points dense workloads toward facilities designed for them rather than a retrofitted comms room.
Efficiency is the lever you control everywhere. Consolidating onto current-generation servers that do more work per watt directly reduces the kilowatts you must buy and cool, which is why a power-aware refresh frequently pays for itself in energy alone. Operating models that share infrastructure efficiently - including consumption-based platforms such as HPE GreenLake - can align cost with actual use and avoid powering idle capacity.
Making the siting decision
Turn it into a sequence. Establish the power and density of what you are deploying; check whether your own site can supply and cool it within a sensible timeframe; if it can, on-prem is on the table; if it cannot, colocation is usually the realistic route to power and cooling at density; and reserve cloud for the elastic, bursty or strategically-outsourced portion. Layer the energy price and grid lead time over the top, because in 2026 those can override the technical preference entirely.
We model this power-first, not cloud-first - the right home for a workload now depends as much on kilowatts and grid access as on software. Start the conversation with our services team, and read the cost framework in cloud vs on-prem TCO in 2026 to put numbers around the choice.