The UK data centre boom: what it means for commercial buildings
The United Kingdom is currently experiencing the fastest growth in data centre construction in its history. AI model training and inference demand, cloud computing expansion and the proliferation of edge computing are driving a wave of investment in hyperscale and co-location data centre capacity across the London metropolitan area, the M25 corridor, and major regional cities. Slough, Wembley, Canning Town, Hayes and the wider Thames Valley are seeing new data centre openings or expansions that collectively represent billions of pounds of investment.
For commercial buildings, this matters because it changes the connectivity calculus. The question for an IT Director or Facilities Director is no longer just “how do we connect devices within the building”, it is “how does the building connect to the data centre infrastructure that hosts our AI compute, cloud services and business-critical applications?” The answer to that question starts with the last-mile connection between the building and the carrier network or dark fibre infrastructure that reaches the data centre.
What the last-mile connection looks like
The last-mile connection is the physical link between a commercial building and the carrier network. For most London offices, this is a fibre circuit delivered by BT Openreach, Virgin Media Business, Zayo, CityFibre or one of a dozen other carriers, terminating at an optical distribution frame (ODF) in the building’s comms room. From the ODF, internal single-mode fibre cabling routes to the core switching infrastructure, which distributes connectivity across the building.
The capacity and resilience of this last-mile connection is the limiting factor for AI application performance in most commercial buildings. An AI application running inference on a GPU cluster in a co-location data centre is only as responsive as the network path between the building and that GPU cluster. If the last-mile connection is saturated by routine business traffic, AI application performance degrades. If the last-mile connection is a single circuit with no resilience, an AI-dependent business is offline when that circuit fails.
For buildings where AI applications are running critical business processes, real-time analytics, AI-driven customer service, AI-assisted financial processing, a single last-mile circuit is not adequate.
Resilient last-mile connectivity, two diverse carrier paths, routed through physically separate infrastructure, terminating at separate entry points in the building, is the correct specification for AI-dependent workloads.
The capacity and resilience of the last-mile connection is the limiting factor for AI application performance in most commercial buildings.
Dark fibre and the direct data centre connection
For organisations running significant AI workloads on co-location data centre infrastructure, the latency of the carrier network path between the building and the data centre can itself become a performance constraint. AI inference applications are latency-sensitive. A 10ms round-trip to a GPU cluster is acceptable; a 50ms round-trip caused by a congested carrier network path is not.
Dark fibre, unlit fibre optic cable leased from a carrier or infrastructure provider, which the customer lights with their own optical transceivers, provides a direct, dedicated, low-latency path between a building and a data centre. It is not the right solution for every organisation, but for those running latency-sensitive AI workloads on co-location infrastructure in the same city, it eliminates the carrier network variability that degrades AI application performance.
The building-side element of a dark fibre connection requires internal single-mode fibre cabling between the building’s entry point and the core switching infrastructure in the comms room. That is a structured cabling project, and the cabling specification for a dark fibre entry is more demanding than for a standard carrier circuit, because the optical performance requirements of single-mode fibre at 10G, 25G or 100G speeds leave no margin for cabling errors.
If the last-mile connection is saturated by routine business traffic, AI application performance degrades regardless of the data centre behind it.
The internal fibre backbone: the overlooked constraint
Most discussions of last-mile connectivity focus on the external circuit. The internal fibre backbone, the cabling between the comms room and telecommunications rooms on each floor, or between multiple comms rooms in a larger building, is often the actual constraint.
A building with a 10G last-mile circuit feeding a multimode fibre backbone installed in 2005 has an internal connectivity constraint that limits the performance of every AI application running on it. Modern AI workloads require OS2 single-mode fibre backbone for runs over 300 metres, OM4 or OM5 multimode for shorter runs. Many existing London office backbones are OM1 or OM2, suitable for 1G Ethernet, inadequate for 10G and above at typical building backbone distances.
An internal fibre backbone upgrade is a structured cabling project that can typically be delivered in a single weekend in most commercial buildings, and it unlocks the full capacity of both the external last-mile connection and the AI applications that depend on it.
The major co-location and hyperscale data centre clusters near London are concentrated in Slough (Microsoft, Amazon, Equinix), Wembley (Ark Data Centres), the Docklands (Equinix LD4-LD8), and Hayes (Google, NTT). The distance between a City of London office and the nearest major data centre is typically 15 to 40km.
At that distance, a correctly specified last-mile circuit on a carrier network delivers sub-10ms round-trip latency to co-location infrastructure, adequate for most AI applications. Dark fibre to the same data centre delivers sub-5ms, the difference that matters for real-time AI inference. The specification choice depends on the latency sensitivity of the workload.
Standards and sources
- Uptime InstituteData centre resilience and standards
- BSI EN 50600 data centre standardData centre facilities and infrastructure standard
Frequently asked questions
What is the last-mile connection and why does it matter for AI applications?
The last-mile connection is the physical link between a commercial building and the carrier network that provides connectivity to the internet, cloud services and co-location data centres. For AI applications running inference on remote GPU clusters, the last-mile connection is the primary determinant of application latency and throughput. A congested or undersized last-mile circuit directly degrades AI application performance. For AI-dependent workloads, a resilient, appropriately sized last-mile connection is a prerequisite, not a nice-to-have.
What is dark fibre and is it appropriate for commercial buildings?
Dark fibre is unlit optical fibre cable leased from a provider, which the customer lights with their own optical transceivers. It provides a direct, dedicated, low-latency path between a building and a data centre, eliminating carrier network variability. Dark fibre is appropriate for organisations running latency-sensitive AI workloads on co-location infrastructure within the same city. It is not the right solution for all organisations. The cost is higher than a standard carrier circuit, and it requires technical capability to manage the optical layer.
What fibre type is required for a 10G or higher internal building backbone?
For building backbone runs over 300 metres, OS2 single-mode fibre is required for 10G and above. For shorter runs, OM4 multimode supports 10G up to 400 metres and 40G up to 150 metres. OM5 multimode supports wavelength division multiplexing applications. OM1 and OM2 multimode, common in buildings cabled before 2010, are adequate for 1G Ethernet but inadequate for 10G at typical building backbone distances. A building backbone upgrade from OM1/OM2 to OM4 or OS2 is typically a one-weekend project and unlocks the full capacity of the external last-mile connection.
How should a commercial building specify resilient last-mile connectivity?
Resilient last-mile connectivity for an AI-dependent building requires two diverse carrier circuits, delivered by different carriers over physically separate infrastructure, entering the building at separate points. The circuits should be terminated on separate routers configured for automatic failover, with a failover time of less than one minute. The internal cabling infrastructure must route both circuits to the comms room independently, with no shared conduit or trunking that could be damaged in a single physical event.