Everyone talks about AI data centres. Almost nobody talks about where AI actually begins.

Refurbishing a hospital while it remains in operation is one of the most demanding infrastructure challenges in construction. Every trade working in a clinical environment must coordinate around patient safety, infection control protocols, noise restrictions, limited working hours, and the fundamental constraint that the building cannot stop functioning while the work takes place.

A hospital network failure is not an inconvenience. It is a clinical event. Patient monitoring systems, infusion pumps, imaging transfer, electronic prescribing, access control, nurse call - the list of IP-connected systems on which patient safety now directly depends has grown to the point where the network is, functionally, clinical infrastructure. Yet the design assumptions underlying most hospital networks were set when the demands on them were a fraction of what they are today.

Most post-mortems on underperforming AI deployments focus on model selection, data quality, or tooling choices. Infrastructure is rarely the headline. It should be.

Everyone's talking about AI agents and model capabilities. But as these workloads move from demos into production, the real bottleneck is becoming clear - it's the physical infrastructure underneath.