AI infrastructure is frequently discussed in terms of what it takes to build capacity. Once that capacity enters production, another challenge begins: operating infrastructure where higher densities can reduce the time available to identify and respond to problems. 

Sessions at Tech Show London examined what that means once AI infrastructure must operate at scale. From testing how systems will behave before they go live, the challenge moves into maintaining higher-density environments and spotting changes quickly enough to act. It also raises a longer-term question: how can operational support scale when skilled technicians cannot simply increase at the same rate as capacity? 

That operational journey starts before the infrastructure is switched on. 

Commissioning Starts Before the Infrastructure is Built 

During the End-to-End Digital Twin for Data Centres – Vision or Reality session at Tech Show London, Stephan Buchholz Solution Director Simulation for Industrial Systems at Siemens, argued for a simulation-first approach, particularly as liquid cooling becomes part of AI infrastructure. 

His proposed process starts by modelling systems during design so cooling components can be sized and control strategies tested. During the build phase, that model can be connected to a simulation of the automation system, allowing control code and sequences of operation to be tested before physical commissioning. 

The model can also support training. 

“You have a great way to train operators of a data centre AI factory on the simulation and digital twin even before the real facility has been built,” said Buchholz in the session End-to-End Digital Twin for Data Centres – Vision or Reality. 

That becomes more relevant as the infrastructure being commissioned changes. 

Higher Density Leaves Less Room to React 

During Seeing Everything, Missing Nothing: Absolute Operational Visibility for High-Density AI Deployments, Ben Gliwa, VP Engineering and Asset Management Operations at Bulk Infrastructure, described how higher-density deployments are bringing customer compute and supporting data centre infrastructure closer together, particularly as liquid cooling extends towards the chip. 

Gliwa said operations could no longer be treated as “set and forget”, with systems needing to adapt as customer designs and requirements change. 

He contrasted higher-density environments with conventional air-cooled deployments, where operators may have had more options to respond temporarily to cooling problems. As density rises, he argued, response needs to become faster. 

Speaking during Agentic AI for Services at Scale: How Agentic AI Transforms Data Centre Services, Natasha Nelson, CTO of Services at Schneider Electric, said technicians are becoming harder to train, hire and retain while the systems they maintain become more complex. 

Nelson described a servicing environment where technicians are being asked to work across more complex systems, while some of the work remains physically demanding or repetitive. 

Her response is greater instrumentation at component level. Nelson described using sensors and telemetry to provide visibility into equipment condition before problems develop into failures. 

That moves more of the maintenance task towards understanding what the infrastructure is telling the operations team before physical intervention is required. 

Monitoring Needs to Reflect What is Happening Now 

Collecting more operational data only helps if it arrives quickly enough to support a response. 

In Finding Order Amongst Chaos, Michael Lawrence of Leviton showed why that visibility cannot treat the data centre as one uniform environment. He divided it into four subsystems — entry point, interconnect, front end, and back end — with different constraints becoming more important in each. 

Those different pressures become particularly important in the GPU-intensive back end, where higher power demand creates a corresponding cooling challenge. Elsewhere, latency or deployment speed may be the more immediate concern. 

For operations teams, that means monitoring needs to reflect what could constrain each part of the infrastructure rather than relying on the same indicators everywhere.  

As part of the panel discussion Automating Operations Beyond the Build, Chris Markstein, Data Centre Lead at Inductive Automation, argued that monitoring needs to move away from intermittent snapshots. 

“If you're relying on intermittent polling to figure out what the status of your robot is every five seconds, you don't have a digital twin.” 

Markstein advocated event-driven systems where changes in equipment state are communicated immediately. Otherwise, he argued, operators can be looking at a historical record rather than the current condition of the environment. 

The Bulk Infrastructure and EkkoSense discussion applied similar thinking to cooling and capacity management. Gliwa described using operational data to support cooling optimisation, hotspot identification and incident response, with predictive modelling intended to give teams earlier warning of developing problems. 

Automation Changes How Support Scales 

The same robotic automation panel raised another constraint: people. 

Speakers argued that growth in data centre capacity cannot simply be matched with equivalent growth in technician numbers. Newer infrastructure also introduces heavier equipment and different handling requirements. 

In Automating Operations Beyond the Build, Jacob Cummings, Vice President of Business Development at PHS West, said automation should therefore be viewed as a way of changing how technicians work. 

“Don't look at it like it's going to be a complete replacement of technicians, it's going to be an enhancement of technicians,” said Cummings. 

As equipment becomes heavier and capacity expands, repeatable physical work places more pressure on the available workforce. Automation can take on some of those tasks, allowing specialist staff to spend more time on work that depends on technical judgement. 

What AI Infrastructure Readiness Now Requires 

Across the Tech Show London sessions, the operational picture starts before the first workload goes live. Simulation can expose problems during design and prepare teams for commissioning. Once infrastructure enters production, higher densities make equipment condition and changes in the environment more important to detect early. 

The discussions also showed how technology and operational expertise must work together. Better monitoring needs to produce information quickly enough for someone to act on it, while automation can relieve pressure on specialist teams without removing the judgement required to operate complex infrastructure. 

The result is a different measure of readiness for AI infrastructure. Delivering the capacity is one part of the job. Operators also need to prove that they can understand how it is behaving, identify when conditions begin to change, and respond at the speed the environment requires.