A data centre expansion boom in the UK is being held back by cable problems that connect AI infrastructure.

TechRadar reported on Monday that UK data centre operators are facing cable supply, latency and design-change issues at the same time as they expand to meet rising AI demand.

The survey was conducted by data centre infrastructure specialist Onnec. It found that 77 percent of UK data centre operators said they face major bottlenecks when procuring cables to support AI workloads. Another 72 percent said competition to invest in AI is creating additional cable problems and that it could take years to resolve them.

The survey also found strains on performance. Some 37 percent said they are experiencing latency or bottlenecks in AI training environments. Another 65 percent said demand for AI processing capacity at data centres is continuing to grow alongside those problems. That suggests the issue is not a temporary supply disruption, but closer to a structural problem where the pace of facility expansion and shifting AI requirements diverge.

Operators also reported that requirements are changing before builds are finished. Some 83 percent said that by the time hardware comes online, AI requirements have already advanced significantly, making upgrades almost immediately necessary to meet new demand and capacity. Another 33 percent said they cannot expand data centres without large-scale cable construction work.

Many also said early design decisions have led to later cost burdens. Some 44 percent said insufficient advance cable planning caused project delays or additional costs. While the focus was on securing power and GPUs, network connectivity was pushed down the priority list, and bottlenecks emerged at the operational stage.

Matt Salter (맷 솔터), Onnec's global data centre head, said everyone is focused on securing power and GPUs, but the network that connects the hardware is too often finalised later. He said once racks are placed and aisles narrow, site teams have little choice but to work around the existing design.

Salter said that even if power and GPUs are ready, they must ultimately be able to communicate with each other, and if the network cannot handle the traffic, AI environments will not deliver the expected performance. He said cables should be included in designs before layouts and aisles are fixed, and that operators need to secure capacity, access and flexibility not only for the first hardware deployment but also for expansion afterwards.

The survey showed that even as the UK increases data centre investment to secure leadership in AI technology, the pace of expansion is not determined solely by securing power or semiconductors. As the bottleneck in the AI infrastructure race shifts to in-room connectivity, the priority in future data centre design is increasingly likely to move toward cables and network capacity.