Why Is AI Making Cooling Companies So Valuable?


NVIDIA sells the chips.

But every more powerful AI chip creates another problem:

Heat.

AI servers consume enormous amounts of electricity. And almost all of that electricity eventually becomes heat that has to go somewhere.

If you can't remove the heat, the expensive GPUs can't keep running.

That means the AI boom isn't creating demand only for chips.

It's also creating demand for the infrastructure required to keep those chips alive.

And some surprisingly old industrial companies are becoming part of the AI economy.

Follow the heat, and another industry appears.

Why Are AI Data Centers Getting So Hot?

A traditional data center contains rows of racks filled with servers.

For years, one of the main ways to cool those servers was relatively simple.

Move cold air through the room.

Let fans push that air through the equipment.

Carry the hot air away.

It worked.

Then AI changed the density of computing.

Instead of spreading computing power across large amounts of space, companies began packing enormous amounts of it into individual racks.

Vertiv's new high-density infrastructure, for example, is designed to support rack densities from 50 kW to more than 100 kW per rack. Schneider Electric is already building cooling equipment for next-generation AI infrastructure at the megawatt scale.

The important thing isn't remembering the numbers.

It's understanding what they mean.

More computing is being packed into less space.

More electricity enters that space.

More heat has to leave it.

And eventually, moving cold air around the room becomes increasingly difficult.

So Liquid Is Moving Closer to the Chip

This is where liquid cooling enters the story.

The basic idea is much easier than the technology sounds.

Instead of relying only on cold air to carry heat away, liquid can move much closer to the hot components and carry that heat out more efficiently.

That requires new infrastructure.

Coolant distribution units.

Chillers.

Pumps.

Heat exchangers.

Valves.

Pipes.

Sensors.

You don't need to remember those names.

They are different pieces of the same problem:

Move enormous amounts of heat away from enormously expensive computers.

And solving that problem is becoming a business of its own.

Vertiv Doesn't Make AI

Vertiv is not a household name.

But inside the data-center industry, it is a major critical-infrastructure company.

Vertiv doesn't make AI models or GPUs.

It provides things data centers need to keep running, including power systems, thermal management and cooling infrastructure.

A simple way to think about it is:

NVIDIA helps provide the brain. Vertiv helps provide the power and cooling that keep the brain running.

And Vertiv is putting real manufacturing capacity behind that demand.

In July 2026, the company announced that it expects to double chiller production capacity at its Tognana campus in Italy by the end of the year.

Vertiv's CEO said AI is creating thermal demands that essentially didn't exist two years earlier.

The company has also been buying more cooling capability.

In June, Vertiv completed its acquisition of ThermoKey, an Italian company specializing in heat-rejection and heat-exchange technology.

That's an important signal.

Heat is no longer just an engineering problem.

It's causing factories to expand and companies to buy other companies.

A Company From 1836 Is Building AI Infrastructure

Then there is Schneider Electric.

Its history goes back to 1836.

That's almost two centuries before ChatGPT.

Schneider grew through heavy industry and later became deeply involved in electrical distribution, automation and energy management.

Today, those old industrial capabilities happen to be extremely useful.

AI data centers need huge amounts of electricity.

That electricity has to be distributed and controlled.

And the resulting heat has to be removed.

Schneider has expanded deeper into data-center liquid cooling, including through Motivair.

In 2026, Motivair by Schneider Electric introduced a coolant distribution unit capable of providing up to 2.5 MW of cooling, with systems designed to scale beyond 10 MW.

There's something fascinating about that.

One of the companies building infrastructure for the AI age was founded almost two centuries before AI existed.

New technologies don't always create only new companies.

Sometimes they give old companies an entirely new reason to matter.

But Modine May Be the More Interesting Story

Modine was founded in 1916.

For more than a century, it has worked on thermal management.

Historically, that meant things such as heat-transfer systems for vehicles, heavy equipment and industrial applications.

It sounds like an old industrial company.

Then look at what happened to its data-center business.

Modine's data-center product sales were:

Fiscal 2024: $294 million

Fiscal 2025: $644 million

Fiscal 2026: $1.11 billion

In two years, that business grew to nearly four times its previous size.

Now compare that with the entire company.

Modine generated approximately:

$3.18 billion in fiscal 2026 sales.

That means data-center products represented roughly:

35% of company sales

on a simple sales comparison.

A business that generated less than $300 million two years earlier had grown to a scale equivalent to more than one-third of the company's annual sales.

That's more than a hot new product.

It's beginning to change what the company is.

Modine Is Changing the Company Around the Opportunity

The numbers get more interesting.

Modine committed $100 million to expanding North American manufacturing capacity for its Airedale data-center cooling business.

That expansion includes manufacturing, engineering, testing and new facilities across several U.S. locations.

Then the company went further.

Effective April 2026, Modine separated Data Centers into its own operating segment.

The company said managing the business independently would help it allocate capital toward the growth opportunity while focusing on margins and cash flow.

And in January, Modine made an even bigger strategic move.

It agreed to combine its Performance Technologies business with Gentherm, leaving the remaining Modine increasingly focused on climate solutions such as data-center cooling and commercial HVAC.

At the time, Modine said it expected its data-center business to grow 50% to 70% annually over the following two years, potentially putting it significantly above an earlier $2 billion fiscal 2028 revenue target.

Think about what has happened.

A company founded in 1916 around thermal engineering is reallocating factories, capital and even parts of its corporate structure around a market being accelerated by AI.

AI didn't turn Modine into an AI company.

It made thermal management much more important.

That's a different kind of AI story.

And the Latest Numbers Show the Demand Hasn't Disappeared

The story didn't stop with fiscal 2026.

In July 2026, Modine reported that its data-center business had recorded three consecutive quarters of record order intake.

Its backlog had nearly doubled over the previous year.

Demand was strong enough that the problem had shifted partly from finding customers to getting enough equipment produced.

The company said near-term supply-chain bottlenecks were affecting its Data Centers segment while it continued spending to expand capacity.

That creates an important distinction.

Strong demand doesn't automatically mean easy profits.

Sometimes demand grows faster than a company can manufacture what customers want.

And that brings us to the number worth watching next.

What Matters Next for Modine?

The obvious number is revenue.

And the revenue growth has been extraordinary.

$294M → $644M → $1.11B

But revenue alone doesn't tell the whole story.

During fiscal 2026, Modine's company-wide gross margin fell from 24.9% to 23.0%.

The company said the decline was driven primarily by temporary costs related to expanding data-center production, along with higher material costs and tariffs.

That's not necessarily bad news.

Building new capacity costs money.

But it tells us what the next question should be.

Not simply:

Can Modine sell more data-center cooling equipment?

It clearly can.

The more interesting question is:

Can it expand production fast enough while improving the economics of that growth?

If capacity catches up with demand and margins improve, that would suggest AI isn't simply increasing Modine's revenue.

It could be improving the quality and scale of the business itself.

If sales keep exploding but production costs and bottlenecks keep absorbing the gains, the story becomes less straightforward.

That's why the next useful numbers aren't only sales.

They're:

Data-center revenue growth

Backlog

Production capacity

and

Margins.

No single one tells the whole story.

Together, they do.

Selling the Cooling System May Be Only the Beginning

There's another reason this business is interesting.

A cooling system isn't necessarily a one-time transaction.

Once the equipment is installed inside a mission-critical data center, it has to keep working.

That can create demand for:

maintenance,

replacement parts,

commissioning,

controls,

monitoring,

upgrades,

and technical support.

Modine's Airedale business already sells a broader set of cooling solutions and services rather than simply individual cooling machines. Its $100 million expansion includes not only manufacturing but also engineering, product development and testing capacity.

Vertiv is moving in a similar direction.

Its infrastructure spans power, cooling and services, and the company is even working with infrastructure-finance providers on integrated power-and-cooling deployments.

So the economic opportunity can extend beyond:

Build equipment → sell equipment.

It can become:

Design → equipment → installation → operation → maintenance → upgrades.

The hotter and more complex AI infrastructure becomes, the more valuable reliability can become.

Then Follow the Supply Chain One Level Deeper

Cooling companies need components too.

Pumps.

Fans.

Motors.

Compressors.

Valves.

Heat exchangers.

Controls.

Sensors.

And if demand for cooling systems rises faster than expected, demand for those components can rise with it.

Modine's latest results already provide a glimpse of this.

The company said it was dealing with near-term supply-chain challenges in its Data Centers business even while orders remained strong.

So follow the money backward:

AI model

↓

GPU

↓

AI server

↓

cooling system

↓

chillers, CDUs and heat exchangers

↓

pumps, motors, valves and components

↓

installation and service

The AI economy spreads much further than the AI companies themselves.

But What If AI Chips Simply Get Cooler?

There is an obvious counterargument.

Semiconductors improve.

AI hardware could perform more calculations using less energy.

Future chips may also tolerate higher operating temperatures.

If that happens, perhaps today's forecasts for cooling infrastructure are too optimistic.

The cooling industry itself is aware of this argument.

When Modine launched a new 3+ MW TurboChill system in 2026, the company explicitly addressed speculation that future chips operating at higher temperatures might reduce the need for chillers.

Modine's response was that real data centers still operate across different climates, different rack densities and varying thermal conditions, so mechanical cooling can remain necessary for reliability.

Of course, Modine sells cooling equipment, so its argument should be viewed in that context.

But there's a broader point.

A more efficient chip does not automatically mean a cooler data center.

Imagine a new AI chip performs twice as much work per unit of energy.

That's a huge improvement.

But what happens if companies install three times as many chips because AI demand keeps growing?

Total power consumption can still rise.

And so can total heat.

The future of cooling therefore depends on a race between two forces:

AI hardware efficiency

and

total growth in AI computing.

That's the same pattern appearing across many parts of AI infrastructure.

Efficiency pushes resource use down.

Scale pushes it back up.

The AI Boom Doesn't End at the Chip

When people think about the economics of AI, they naturally look at companies like NVIDIA.

That makes sense.

The GPU is essential.

But a GPU needs electricity.

Electricity becomes heat.

Heat needs cooling.

Cooling systems need pumps, chillers, heat exchangers, controls and other components.

Then someone has to install and maintain all of it.

One piece of AI infrastructure creates demand several layers down the supply chain.

That's why one useful way to understand a new technology isn't simply to ask:

Who makes the technology?

Ask instead:

What new problems does the technology create?

If AI needs more electricity, who solves that problem?

If AI creates more heat, who solves that problem?

If AI needs more water, networking or security, who solves those problems?

Sometimes the most interesting businesses around a new technology aren't making the technology itself.

They're solving the problems created by its success.

Modine is a good example.

Its data-center sales went from $294 million to $1.11 billion in two years.

It committed $100 million to additional manufacturing capacity.

It created a separate Data Centers operating segment.

And now its backlog has nearly doubled in a year.

But the next question is no longer whether demand exists.

It's whether Modine — and companies like it — can convert that demand into capacity, margins and durable economics.

That's where the story becomes more interesting.

The AI boom doesn't end at the chip.

Every watt that enters an AI server eventually becomes heat.

Follow the heat, and another industry appears.

BEYOND THE OBVIOUS.

Sources