AI Infrastructure Is Testing the Limits of Data Centre Operations
2 days ago
Higher-density AI infrastructure is changing the demands placed on data centre operators. Insights from Tech Show London and Frankfurt examine why commissioning, cooling and operational resilience need greater attention as AI capacity expands.
Investment in AI infrastructure is creating new operational challenges for data centres, where higher-density computing systems require more demanding cooling arrangements and closer monitoring of equipment conditions.
While much of the attention surrounding AI infrastructure has focused on the construction of new facilities and the availability of electricity, operators must also consider how that capacity will perform once it enters production.
According to Uptime Institute, the vast majority of data centre racks remain below 10kW, while some high-density AI systems exceed 40kW and emerging configurations can surpass 100kW.
These higher densities can reduce the time available to identify and respond to equipment failures, particularly when cooling systems are closely integrated with computing hardware.
Research and practitioner discussions from Tech Show London and Tech Show Frankfurt examined how these requirements are changing the way AI infrastructure is operated.
AI Infrastructure Requires Different Operating Conditions
A Tech Show Frankfurt Insights analysis examined the technical requirements of AI infrastructure, where GPU clusters rely on high-speed networking and storage systems to support demanding workloads.
The concentration of computing equipment creates additional pressure on facility infrastructure, particularly where existing data centres were designed for lower-density deployments.
Cooling is one of the clearest examples.
Uptime Institute identified conventional perimeter air cooling as typically supporting approximately 20–25kW per rack, while close-coupled air systems can accommodate higher densities.
For some AI deployments, direct liquid cooling provides an alternative by transferring heat away from processors through circulating coolant.
However, these systems introduce additional operational requirements. Pumps, coolant distribution units and leak detection systems must be maintained alongside conventional facility infrastructure.
The Frankfurt analysis also examined how organisations can assess whether colocation facilities are suitable for AI workloads, including the importance of verifying available power and cooling capacity before installation.
Higher Density Changes the Operational Risk
The operational consequences of higher-density infrastructure were also discussed at Tech Show London 2026.
Ben Gliwa, VP Engineering and Asset Management Operations at Bulk Infrastructure, argued that high-density deployments could no longer be treated as “set and forget”.
His comments reflected the demands placed on operators as cooling systems become more closely integrated with computing equipment.
In conventional environments, operators may have more time or alternative measures available when cooling conditions change. Higher-density systems can narrow those options, increasing the importance of identifying faults before they affect production workloads.
The challenge extends to how organisations monitor infrastructure and prepare staff to respond when conditions change.
Discussions at Tech Show London examined whether traditional monitoring arrangements could provide information quickly enough to support operational decisions, alongside the potential for simulation and automation to improve maintenance practices.
The full Tech Show London Insights analysis explores how operators are approaching commissioning, operational visibility and workforce pressures as AI infrastructure scales.
Operational Readiness Goes Beyond Installed Capacity
The two regional perspectives raise questions about how organisations assess whether new AI infrastructure is ready for production.
An installation may meet its planned computing capacity while still facing constraints in cooling performance or maintenance capability.
For operators, those considerations begin before deployment. Cooling arrangements need to accommodate the proposed hardware, while monitoring and recovery procedures must reflect the operating conditions of the facility.
Responsibilities also need to be clear where customers operate their own equipment within colocation environments. Problems affecting cooling or power infrastructure can cross the boundaries between facility operators and tenants.
The Frankfurt analysis established the technical dependencies supporting AI workloads, while the London discussions examined how those systems can be managed once operational.
For enterprises investing in AI infrastructure, assessing capacity therefore requires more than confirming the number of GPUs installed or the power available at a site.
The ability to operate that infrastructure reliably, maintain equipment and respond to changing conditions will determine how much of the installed capacity can be delivered as a dependable service.
