Nick Maggs, Managing Director Hard Services at OCS, makes the case that AI’s greatest data centre threat isn’t computing power, it’s the ageing cooling infrastructure quietly running underneath it.
Much of the discussion around Artificial Intelligence and data centres places focus on computing power. We hear about new processors, larger servers and facilities designed to handle far more information than ever before. What receives far less attention, however, is the physical infrastructure that allows all this equipment to operate safely, particularly cooling.
For many data centres, the greatest challenge is not the arrival of AI in recent years, but the condition of the systems already in place. Heating, ventilation and air conditioning equipment installed years ago is now being asked to support workloads it was never designed to carry. These systems may still be operating, but the margin for error has narrowed considerably.
AI has not created a new problem but has instead brought existing weaknesses to the surface, exposing the limits of ageing plant that has been kept running through repair rather than replacement.
What has changed inside data centres
To understand why cooling is under strain, it helps to look at what modern data centres now contain. AI workloads rely on powerful processors that perform vast numbers of calculations at once, generating a great deal of heat along the way. To make best use of space, they are often installed close together inside tall metal frames known as racks.
As more equipment is fitted into each rack, heat output rises. At the same time, AI workloads tend to run continuously rather than switching on and off. Unlike traditional computing, where demand often fluctuated, AI systems can operate at high intensity around the clock.
This creates a steady and sustained thermal load. Cooling systems must remove heat consistently, without respite. For newer facilities designed with this in mind, that challenge can be managed. For older sites, it places existing plant under constant pressure.
Cooling systems designed for a different era
Many data centres still rely on cooling equipment that was installed when computing demands looked very different. These systems were built to cope with lower heat output, more variation in demand and regular opportunities for downtime.
Over time, components such as chillers, air handling units and control systems reach the end of their intended working life. Rather than being replaced, they are often repaired and kept running. Each repair makes sense in isolation; full replacement is disruptive and costly and budgets are rarely generous.
The issue, however, is cumulative. Equipment that has been repaired repeatedly becomes less reliable. Parts wear unevenly and performance diminishes. Under lighter workloads, these weaknesses can remain hidden. Under sustained AI demand, they become far more apparent.
When maintenance becomes a warning sign
Ageing infrastructure rarely fails without warning. More often, it shows its age through patterns. Engineers are called out more frequently, the same components require repeated attention and temperatures fluctuate more than they should.
This growing volume of reactive maintenance is often treated as business as usual. In reality, it is an early warning sign that shouldn’t be taken lightly. It indicates systems that are no longer suited to the role they are being asked to play and risk disruptive consequences as a result.
In a data centre environment, cooling problems escalate quickly. If temperature control is lost, equipment can overheat within minutes. Systems may shut down automatically to protect themselves. Services can be interrupted, sometimes without warning.
For organisations relying on AI-driven services, this level of risk is increasingly difficult to tolerate. Over time, this pattern turns maintenance from a solution into a signal that a different approach is needed.
Why lifecycle planning matters
Managing assets over their full lifespan, rather than reacting when they fail, is essential for minimising disruption to the functioning of critical assets. Lifecycle planning means understanding how long equipment is expected to last, how its performance changes over time and when replacement becomes the sensible option. This approach allows organisations to plan investment rather than firefight breakdowns. It also supports better decision-making, particularly when workloads are evolving quickly.
In data centres supporting AI, lifecycle planning for cooling equipment is becoming essential. Systems cannot remain static while computing demand increases around them. Without a clear replacement strategy, operators risk being forced into urgent decisions at the worst possible moment.
Using existing data more effectively
Modern cooling systems generate large volumes of data through building management systems. Temperatures, run times and fault histories are recorded constantly. This information offers valuable insight into how assets are performing.
When reviewed properly, trends such as rising energy use, increasing alarm frequency or repeated repairs point to declining asset health. Acting on these signals allows upgrades and replacements to be planned calmly, rather than under pressure.
The challenge lies in using this data to support long term decisions rather than treating it as a compliance exercise.
Treating cooling as critical infrastructure
As AI continues to shape data centre demand, cooling systems will carry more responsibility than ever before. Older equipment that once coped adequately is now being pushed closer to its limits, often without a corresponding change in how it is managed or funded.
The operators best placed to support next-generation workloads will be those who recognise the risk posed by ageing assets and address it directly. That means moving away from short-term repairs, investing in lifecycle planning and treating cooling with the same seriousness as power and connectivity.
AI has raised expectations for what data centres can deliver. Meeting those expectations depends as much on the condition of existing infrastructure as it does on the technology being installed today.
