AI datacentre Cooling

Cooling: The Next Constraint in AI Infrastructure (And Why It’s Closer Than You Think)

After more than 20 years delivering infrastructure at scale, I can tell you this:

Systems don’t usually fail because of the obvious constraint.

They fail because of the one no one planned for.

In my previous article I focused on power.

But the next constraint is already forming.

And it’s cooling.

What People Think

Many discussions around AI infrastructure focus on:

And increasingly, energy availability. Which makes sense.

Because AI workloads are driving unprecedented power demand.

But there’s a second-order effect that’s often overlooked.

Because every watt of power consumed, it turns into heat.

What’s Actually Happening

AI workloads are pushing data centres into a new operating range.

We’re now seeing:

·         Higher rack densities

·         More concentrated compute clusters

·         Continuous, high-intensity processing

(And just to be clear—when I say “compute,” I mean processing power at scale: thousands of specialised processors running continuously.)

That creates one unavoidable outcome:

Heat. A lot of it.

Where It Breaks

On paper, cooling sounds manageable:

“Just upgrade the cooling system.”

In reality, this is where things get complicated.

1. Thermal Density Is Increasing Faster Than Cooling Capability

Traditional data centre cooling was designed for:

·         Lower power densities

·         Distributed workloads

AI flips that.

Now you have:

·         Highly concentrated heat zones

·         Continuous load (no downtime cycles)

Cooling systems weren’t built for this level of intensity.

2. Cooling Requires Infrastructure Too

Cooling isn’t just equipment.

It depends on:

·         Water availability

·         Power (again)

·         Physical space

·         Environmental approvals

So now you have a dependency chain:

More compute = more power = more heat = more cooling = more infrastructure

3. Location Constraints Start to Compound

This connects directly to my last article.

If data centres are moving closer to power.

They may also be moving to locations where:

·         Water is limited

·         Climate conditions are harsher

·         Infrastructure support is thinner

So solving one constraint creates another.

Operator Insight: This Is Where Systems Start to Strain

I’ve seen this pattern before in infrastructure programs.

You solve the primary constraint.

And then secondary constraints start to emerge.

Not because they weren’t there…

But because they weren’t limiting, until now.

Cooling is exactly that.

It’s been part of the system.

But not the limiting factor.

Until now.

The Hidden Complexity Most People Miss

Cooling at scale isn’t linear.

You don’t just “add more cooling.”

You have to rethink:

·         Facility design

·         Equipment layout

·         Airflow or liquid cooling systems

·         Operational models

And most importantly:

How all of this integrates with power and network infrastructure.

The 5 Constraints Are Now Interacting

This is where your framework becomes powerful.

Cooling isn’t a standalone issue.

It sits across:

·         Power - drives heat generation

·         Fibre - drives data movement and workload distribution

·         Land - impacts space and environmental constraints

·         Workforce - specialised skills required

·         Capital - rising cost of advanced cooling solutions

This is no longer a single constraint problem.

It’s a system interaction problem.

What This Means for Industry

We’re already seeing early signals:

·         Increased interest in liquid cooling technologies

·         Data centre design shifting toward higher-density capability

·         More scrutiny on water usage and environmental impact

And importantly:

Cooling is starting to influence design decisions.

Not just support them.

What Happens Next

Over the next few years, expect:

·         Cooling to become a primary design planning consideration

·         New technologies (liquid, immersion) to scale rapidly

·         Increased regulatory attention (especially around water use)

·         Location strategy influenced by both power and cooling capability

And potentially:

Cooling becoming the constraint that slows down deployment.

Even when power is available.

Final Thought

In infrastructure, the first constraint gets all the attention.

The second constraint causes the real delays.

Right now, power is the focus.

But cooling is right behind it.

And if it’s not planned properly

It won’t matter how much compute you have.

Because you won’t be able to run it.

References & Further Reading

Footnote
This article is part of a series exploring the physical infrastructure behind the AI economy.

Disclaimer
The views expressed in this article are my own and are intended for general information and discussion purposes only. They do not represent the views of any employer, organisation, or client.

© 2026 Rodney Terry – Digital Backbone. All rights reserved.