Why Aren't All AI Data Centers Built Somewhere Cold?
AI has a heat problem.
Thousands of GPUs running around the clock generate enormous amounts of heat. That heat has to go somewhere, which means modern AI data centers need increasingly sophisticated cooling systems.
So there is an obvious question:
Why not build them somewhere cold?
Alaska. Iceland. Northern Canada. Finland.
If cooling is such a problem, putting servers somewhere that is cold for much of the year sounds almost too obvious.
And here's the interesting part.
The idea isn't wrong.
Cold climates can help data centers reject heat more efficiently. Weather conditions, including temperature and humidity, can affect cooling design, energy efficiency and water use.
Cold is useful.
But for an AI data center, something else can be even more valuable.
Power.
AI Data Centers Are Becoming Power Projects
Imagine finding the perfect piece of land.
It's cold.
It's large.
It's relatively cheap.
But there is one problem:
You can't get enough electricity to it.
For a large AI campus, that can kill the project.
A modern data center doesn't simply plug into a wall. Electricity may have to travel through generation, transmission infrastructure, substations, transformers and distribution equipment before it reaches thousands of servers.
And AI is pushing those requirements much higher.
Power availability and grid capacity are now primary considerations in data center site selection. In some regions, reliable access to high-capacity electricity has become the dominant factor determining where large facilities can be built.
OVIQQ previously looked at just how much electricity a large AI data center can consume.
That leads to a simple way of thinking about AI infrastructure:
Cold air is useful. Power is essential.
Having Electricity Nearby Doesn't Mean You Can Get It Tomorrow
There is another problem hiding behind the power grid.
Time.
A massive data center may require new substations, transmission upgrades, transformers and other equipment before it can receive the electricity it needs.
That demand is also colliding with another infrastructure problem: a global shortage of large power transformers.
And getting connected can take years.
ASHRAE cites constrained parts of Northern Virginia where utility interconnection timelines can stretch three to five years or longer.
For emerging Texas data center markets, large projects can face interconnection timelines of roughly two to four years, depending on the location and size of the load.
That changes the economics.
Imagine two locations.
Site A
Very cold.
Cheap land.
Excellent natural cooling.
But power takes five years to secure.
Site B
Warmer.
More expensive cooling.
But massive electricity capacity can be secured sooner.
Site B can still be more valuable.
Because for a company spending billions of dollars on AI infrastructure, waiting years for electricity means expensive computing equipment isn't producing revenue.
In this business, speed to power can become speed to revenue.
AI Also Needs an Internet Highway
Electricity alone isn't enough.
A data center has to move enormous amounts of information between servers, networks, businesses and users.
That requires high-capacity fiber connectivity.
And distance can matter.
For latency-sensitive services, being reasonably close to users and major network hubs remains important. But AI is beginning to change the balance.
AI and high-performance computing workloads are pushing site-selection priorities toward power availability, scalable land and long-term infrastructure capacity, even while connectivity remains essential.
So the ideal location isn't simply:
somewhere cold.
It's somewhere where several scarce resources meet.
Power + network + land + cooling + water + permits + expansion capacity.
Finding all of them in one place is much harder.
Then Why Does the Nordic Model Work?
This is where the cold-climate idea becomes genuinely interesting.
Northern Europe doesn't offer only cold weather.
Parts of the region can combine favorable climate conditions with low-carbon electricity, land, connectivity and established data center infrastructure.
One company built around that opportunity is atNorth, a Nordic high-density data center operator with operations across five countries.
And in February 2026, something significant happened.
Equinix and CPP Investments agreed to acquire atNorth at an enterprise value of approximately $4 billion.
CPP Investments is expected to own roughly 60%, with Equinix owning roughly 40%, subject to customary closing conditions and regulatory approvals.
Why pay billions for a Nordic data center company?
Equinix specifically pointed to growing demand from enterprise, AI and hyperscale customers and atNorth's Nordic capacity.
The valuable asset isn't simply cold weather.
It's the combination:
**Climate
- Power
- Land
- Connectivity
- Existing data center capacity**
Cold becomes valuable when the rest of the infrastructure is already there.
Cold Weather Doesn't Eliminate Cooling
There's another misconception worth clearing up.
If the outside air is cold, why do AI data centers still need sophisticated cooling equipment?
Because there are really two problems.
First, heat has to be removed from the chips.
Then that heat has to be rejected from the facility.
Cold outdoor conditions can help with the second problem.
They don't make the first one disappear.
AI racks are becoming so power-dense that traditional air cooling is reaching practical limits for some deployments.
Direct-to-chip liquid cooling has become an important approach for high-density AI and high-performance computing.
In simple terms:
Cold weather helps the building get rid of heat.
But you still need technology to move that heat away from thousands of extremely hot chips.
And that's where another group of companies enters the AI story.
Who Makes Money From This?
Most investors looking at AI start with NVIDIA.
That makes sense.
GPUs are at the center of the computing boom.
But a GPU sitting in a warehouse produces nothing.
It needs electricity.
It needs cooling.
It needs power distribution.
It needs backup systems.
It needs networking.
And it needs somewhere to live.
That creates an enormous physical supply chain behind the digital AI economy.
Vertiv
Vertiv provides power and thermal infrastructure used in data centers, including systems designed for increasingly dense AI computing environments.
One way to think about the business is:
NVIDIA helps build the brain. Companies like Vertiv help keep that brain powered and cool.
Schneider Electric
Schneider Electric operates across electrical distribution, power management and data center infrastructure, including liquid-cooling systems.
This matters because AI data centers increasingly have to treat power and cooling as one integrated engineering problem, rather than two separate systems.
Equinix and atNorth
Equinix represents another layer of the AI economy.
The company isn't designing the GPU.
It's helping provide the physical infrastructure where computing can happen.
Its planned investment in atNorth highlights a different kind of scarcity:
places where large amounts of computing can actually be deployed.
The $4 billion transaction suggests that infrastructure capable of combining power, connectivity, cooling and expansion capacity can itself become a valuable AI asset.
The AI Bottleneck May Not Be the Chip
Now return to the original question.
Why don't we simply build every AI data center somewhere cold?
Because a developer isn't asking only:
Is it cold here?
They're asking:
Can we get hundreds of megawatts of electricity?
How quickly?
Is there a substation nearby?
Can the grid handle us?
Is there enough fiber?
Can we cool the equipment?
Is water available?
Can we get permits?
Can we expand later?
Can we find workers?
Will the local community accept the project?
Power, cooling, water, connectivity, land, regulation, resilience and workforce are all interconnected parts of the site-selection problem.
So the best location isn't necessarily the coldest place.
It isn't necessarily the cheapest land.
And it isn't necessarily the biggest city.
It's the place where enough of the required infrastructure can come together at the right time.
AI Is More Physical Than It Looks
From the outside, AI looks almost entirely digital.
Models.
Software.
Algorithms.
GPUs.
But follow the money one layer deeper and a very physical economy appears.
Electricity.
Transformers.
Copper.
Cooling equipment.
Backup power.
Fiber.
Water.
Land.
AI data centers require enormous amounts of power and cooling, and in some regions deployment is increasingly constrained by the capacity of the electrical grid.
That changes the question for anyone trying to understand where the AI economy is going.
Instead of asking only:
What's the next AI chip?
It may also be worth asking:
What does the world need to run hundreds of thousands of those chips?
Follow that question and companies that don't look like AI companies suddenly appear.
Electrical-equipment makers.
Cooling companies.
Generator manufacturers.
Utilities.
Data center operators.
Infrastructure developers.
The AI race may ultimately depend on something much less glamorous than another breakthrough algorithm.
Finding enough places to plug it all in.
BEYOND THE OBVIOUS.
Sources
ASHRAE — AI Data Center Framework: Site Planning
ASHRAE — AI Data Center Framework: Integrated Design Principles
ASHRAE — AI Data Center Framework: Energy and Thermal Efficiency
Equinix — CPP Investments and Equinix to Acquire atNorth for US$4 Billion