London's power constraint: what data-centre demand means for technology businesses

London's power constraint does not mean every technology business should move its computing or office estate elsewhere. It means that power availability can no longer be treated as an item to confirm after a preferred building, colocation hall or dedicated data-centre site has been selected. A buyer should test connection status, usable capacity, delivery date, resilience, contract exposure and expansion rights before treating any site as viable.

For most businesses, the practical choice is between public cloud, colocation and dedicated infrastructure, not between building a data centre in London and building one elsewhere. The right answer depends on workload latency, data location, resilience, security, procurement leverage and the cost of failure. A London address is useful only when it improves those outcomes more than it increases dependency on constrained infrastructure.

A connection offer is not the same as usable power

Electricity reaches a data centre through several layers of network and site infrastructure. A building may be near a substation but still lack an available connection of the required size. A developer may hold a connection offer without having completed the planning, land, equipment or financial steps needed to energise the site. A prospective customer should therefore ask for evidence of current energised capacity, contracted capacity, the party holding the connection agreement and the conditions attached to future expansion.

The Mayor of London's June 2026 answer on data-centre capacity cited around 760 MW of existing London capacity and more than 8 GW in the connections queue. Those figures describe different things. Existing operational capacity is not necessarily available to a new customer, while queued demand is not a forecast of what will be built. A procurement paper should keep operational, contracted, proposed and queued capacity in separate columns.

Great Britain's connection process is also changing. The National Energy System Operator's connections-reform material explains the move towards a delivery pipeline that prioritises projects able to meet readiness and strategic criteria. In July 2026, Ofgem consulted on measures aimed at speculative data-centre demand, including progress milestones and a proposed commitment fee. These reforms may improve the quality of the queue, but they do not create local network capacity instantly.

Planning status can change the delivery risk

Electricity is only one dependency. A dedicated facility also needs planning permission, land control, cooling, water strategy, network connectivity, backup arrangements and a workable construction programme. The Draft London Plan published for consultation in 2026 introduces a standalone approach to data centres and acknowledges uncertainty over future demand. It remains a draft until the statutory process is complete, so a buyer should not treat it as adopted policy.

For an existing colocation hall, ask whether the provider's quoted capacity is already energised and reserved for the proposed contract. For a facility under construction, ask which permissions are final, which equipment has been ordered and which milestones allow either party to terminate. For a new corporate site, obtain written advice from the distribution network operator before signing a lease whose intended use depends on a large power upgrade.

This is also why the planned article on where London can still make and move things matters. Data centres compete with logistics, production, housing and other infrastructure for land and power. Planning consent for one use does not prove that the wider utilities package is ready.

Cloud, colocation and dedicated capacity solve different problems

Public cloud transfers much of the facility and power problem to a provider. It can offer rapid deployment across more than one availability zone, but it introduces supplier concentration, data-transfer charges, service architecture requirements and contractual limits. The business still needs to know where services and backups run, which failures the provider covers and which resilience work remains with the customer.

Colocation gives the customer more control over equipment and network design while sharing a facility's power, cooling and physical security. A buyer should compare the committed power model, metering, price-escalation clauses, remote-hands service, cross-connect charges, maintenance windows and expansion rights. Cheap initial rack space is of limited value if the provider cannot reserve adjacent capacity for the next stage of growth.

Dedicated infrastructure provides the most control and the greatest project exposure. It may be justified for unusually large, predictable or regulated workloads, but it requires a credible capacity plan rather than a prestige address. The commercial case should include design, connection, equipment lead times, finance, maintenance, staffing and the residual value of specialised property.

Latency and resilience must be measured against the real workload

London proximity matters for workloads where milliseconds affect the service, for private connectivity to financial or media counterparties, or for operational teams that must reach equipment quickly. It matters less when an application can use content delivery, caching or distributed architecture. The buyer should measure end-to-end application performance from the users and counterparties that matter, rather than use the distance between two postcodes as a proxy.

Resilience also needs a failure model. Two halls in the same power or fibre corridor may not provide meaningful geographic diversity. A design should document grid supply, substations, generators, fuel, cooling, carrier routes and operational control. It should then test common-cause failures, not simply count facilities. The government's decision to designate data centres as critical national infrastructure increases the policy attention on resilience, but it does not replace customer due diligence.

Contract terms determine who carries energy and delay risk

A data-centre proposal should state whether electricity is included, passed through or indexed, and how losses, taxes and renewable-energy products are treated. The customer needs an auditable baseline for contracted power, actual consumption and peak demand. It should also understand whether the provider can reduce capacity, move equipment or change the energy-pricing method.

Future capacity deserves its own schedule. Record the amount, delivery conditions, reservation payment, expiry, remedies and evidence required at each milestone. If an expansion depends on a network upgrade, planning decision or new substation, label it as conditional. Do not price a product launch as if conditional power were operational.

Environmental statements require the same discipline. Renewable certificates, direct power contracts, on-site generation and hourly matched supply are different claims. A buyer should ask for the reporting boundary, method, evidence period and treatment of backup fuel before repeating a provider's carbon statement. The UK government's greenhouse-gas conversion factors can support consistent reporting, but they do not validate a supplier's marketing claim.

Another UK region can be the better compute location

A location outside London may offer a clearer connection path, more land and a different cost base. It may also increase latency to London users, reduce carrier choice or create travel and skills constraints. The correct comparison uses the same workload, resilience standard, contract term and delivery date in every location.

Large training workloads, batch processing, archives and secondary recovery may tolerate greater distance. Customer-facing systems, trading connections and some production media workflows may not. A split architecture can keep latency-sensitive services close to London while placing flexible compute elsewhere. The forthcoming Shoreditch Talk guide to industrial climate-technology locations applies the same principle: physical infrastructure should follow operational needs, not the headquarters address.

A practical buyer checklist

Before approving a location or supplier, obtain:

1. Evidence of energised and reserved capacity.
2. The connection holder, network level and expansion conditions.
3. Planning, land and construction status.
4. A workload-specific latency test.
5. Power, cooling, carrier and operational failure maps.
6. Energy-price, capacity and termination clauses.
7. Data-location and subprocessor records.
8. An environmental reporting method with defined boundaries.
9. A comparable option outside London.
10. A scenario showing the impact of a 12-month capacity delay.

Technology teams assessing research-intensive compute should also read Can London turn quantum research into paying customers?, where access to specialised computing is one dependency among several on the route to commercial work.

Limitations

Electricity constraints are highly site-specific. London-wide queue figures cannot establish the capacity or delivery date at one substation, building or data centre. Commercial connection agreements, utilisation and colocation pricing are usually private. This article therefore provides a procurement method, not a capacity forecast or engineering assessment. The regulatory and planning material was checked on 13 October 2026 and must be checked again before publication or a transaction.