Why Build a $15 Billion AI Data Center Next to a Power Plant?
When people think about artificial intelligence, they usually think about:
ChatGPT.
GPUs.
Robots.
Software.
But one of Japan's biggest planned AI infrastructure projects is being built around something much older.
A power plant.
In October 2026, Japan's largest power generator, JERA, Dell Technologies and UK-based RHAELM announced an agreement to develop large-scale AI infrastructure in Japan.
Their first planned project is in Chiba, near Tokyo.
The numbers are enormous.
More than $15 billion.
Up to 400 megawatts of power capacity.
Operations targeted to begin around:
2028.
And the data center will be located at JERA's existing:
Chiba Thermal Power Station.
Why put one of Japan's largest planned AI projects next to a power plant?
The answer is surprisingly simple.
AI needs electricity.
A lot of it.
First, What Is a Data Center?
Imagine thousands of powerful computers inside a giant building.
Those computers store information, run software and perform calculations.
That is basically what a data center does.
AI requires particularly powerful computers.
Those computers need electricity.
But that is only the beginning.
They also create enormous amounts of heat.
So a large AI data center needs:
computers,
electricity,
cooling,
electrical equipment,
buildings,
network connections,
and backup systems.
You cannot simply buy thousands of AI chips and turn them on.
You need an entire factory around them.
This One Could Cost More Than $15 Billion
According to JERA, total capital deployment across all phases of the Chiba project is expected to exceed:
$15 billion
or roughly:
¥2.3 trillion.
That does not mean someone is spending $15 billion just on servers.
The investment is expected to include:
land,
power infrastructure,
facility construction,
and AI computing equipment.
The project is planned to have access to up to:
400 MW
of power capacity.
That does not mean the facility will constantly consume exactly 400 MW.
Think of 400 MW as the maximum planned power capacity available to support the project.
Still, it is an enormous number.
How Big Is 400 MW?
JERA's Chiba Thermal Power Station has a maximum generating capacity of about:
4,380 MW.
The planned AI project would have access to up to:
400 MW.
As a simple capacity comparison:
400 ÷ 4,380 ≈ 9%
Again, this does not mean the data center will constantly consume 9% of the power plant's output.
It is only a comparison between two maximum capacity figures.
But it helps show the scale.
This is not like plugging a computer into a wall.
This is industrial infrastructure.
So Why Build It Next to the Power Plant?
Imagine you build a giant factory.
The factory is finished.
The machines are ready.
The workers are ready.
But there is one problem.
You cannot get enough electricity.
The factory is almost useless.
AI data centers can face the same problem.
Building the computers is not enough.
Someone must deliver enormous amounts of reliable electricity to them.
And connecting a huge new facility to the normal power grid can take years.
So the Chiba project is trying something different.
It is being built beside a place where enormous amounts of electricity are already produced.
The Important Phrase Is "Behind the Meter"
JERA describes the project as:
behind-the-meter.
It sounds complicated.
The basic idea is simple.
A conventional setup might look like this:
Power plant → electricity grid → data center
But the Chiba project is designed to use JERA's existing power infrastructure at the site, reducing dependence on the normal process of waiting for a large new grid connection.
Think about water.
Imagine your factory needs an enormous amount of water.
You could build the factory somewhere else and wait for the city to install much larger pipes.
Or:
build the factory beside the water source.
That is roughly the economic idea.
JERA says this approach could make power capacity available:
years earlier
than a conventional grid-connected development schedule.
That word matters.
Years.
The Power Plant Is Selling Time
Suppose two companies own exactly the same AI computers.
Company A can turn them on in:
2028.
Company B has to wait until:
2031.
Which company can start selling computing first?
Company A.
Which company can start training models first?
Company A.
Which company's expensive GPUs can begin earning money first?
Company A.
This changes the way we should think about electricity.
Electricity is not just another monthly bill.
In AI infrastructure:
electricity can also be time.
And time has economic value.
A GPU sitting in a box does not generate AI computing revenue.
A server without power cannot serve a customer.
A finished data center waiting for electricity is an expensive building full of machines that cannot do their job.
So building beside a power plant can potentially buy something more valuable than cheaper electricity.
It can buy:
years.
Imagine $15 Billion Waiting for Electricity
Now think about the size of the project again.
More than:
$15 billion
of capital is expected to be deployed across its different parts.
Imagine spending billions on:
land,
buildings,
cooling,
electrical equipment,
and computers.
Then imagine waiting years for enough electricity.
The problem becomes obvious.
The cost is not only the electricity bill.
The cost can also be:
waiting.
A giant AI facility that cannot turn on is not really a computing business yet.
It is capital waiting for power.
That is why land beside existing energy infrastructure can become unusually valuable.
The land is not valuable only because of where it is.
It is valuable because of:
what can be turned on there.
The Power Deal Could Last 15 to 25 Years
Reuters reported another important number.
JERA is expected to provide:
400 MW of power capacity
for:
15 to 25 years.
That tells us something important about giant AI data centers.
This is not a temporary electricity problem.
Companies planning facilities like this need to think about power for decades.
Once billions of dollars have been invested in a site, those computers need reliable electricity year after year.
So choosing where to build an AI data center is partly a technology decision.
But it is also:
a long-term energy decision.
This Is Why a Power Company Is in an AI Deal
At first, the partners may look unusual.
JERA.
Dell Technologies.
RHAELM.
Apollo Global Management is also expected to participate as a strategic investment and financing partner for RHAELM.
But each company controls a different part of the system.
JERA brings:
land and power.
Dell brings:
AI computing infrastructure.
RHAELM is expected to develop, construct, operate and finance the data center.
Apollo plans to participate on the investment and financing side.
Put them together and you get something much larger than a room full of GPUs.
You get an industrial system.
AI Starts Long Before the GPU
People often imagine the AI supply chain like this:
GPU → Server → AI
But the real chain can begin much earlier.
For a project like this, a simplified version looks more like:
LNG → Power plant → Electricity → Data center → Cooling → Servers → GPUs → AI
Look at how far the chain extends before reaching the chip.
Fuel has to be purchased.
Electricity has to be generated.
Electrical infrastructure has to exist.
Buildings have to be constructed.
Computers have to be cooled.
And only then can electricity reach the GPUs.
AI may feel like software.
At enormous scale, it starts looking like:
heavy industry.
A GPU Without Electricity Is Just an Expensive Box
Imagine buying the world's most powerful refrigerator.
But your house has no electricity.
How useful is it?
Not very.
The same basic rule applies to AI hardware.
A company can buy advanced GPUs.
It can build sophisticated AI models.
It can spend billions of dollars on servers.
But:
If the computer cannot get electricity, it does not matter how smart the AI is.
This is why the AI boom increasingly involves companies that do not look like AI companies.
Power generators.
Gas suppliers.
Utilities.
Construction companies.
Cooling companies.
Data center developers.
Infrastructure investors.
They all sit underneath the final AI product.
The Best AI Real Estate May Be Where the Power Already Is
For an ordinary office, a company might ask:
Is it close to customers?
Is transportation convenient?
Is the neighborhood attractive?
What is the rent?
A giant AI data center asks different questions.
How much electricity is available?
How quickly can we get it?
Can we get it reliably?
Can we cool the computers?
Can we expand later?
The best piece of land may not be the prettiest land.
It may be land beside:
a power plant.
Because if power becomes scarce, land with fast access to hundreds of megawatts becomes more than real estate.
It becomes part of the computing infrastructure itself.
JERA Wants to Repeat the Model
The Chiba project is not being presented only as one giant data center.
JERA, Dell and RHAELM say they want to create a:
standardized, repeatable model.
The reason is easy to understand.
Every giant AI project has to solve the same basic problems:
Where does the electricity come from?
How does it reach the computers?
How are the computers cooled?
Who builds the facility?
Who pays for it?
If one model works, the companies may be able to use it again.
The partners say they want to explore other JERA sites and eventually support:
multi-gigawatt
AI infrastructure capacity across Japan in the 2030s.
One gigawatt equals:
1,000 MW.
So several gigawatts would mean several times the scale of the planned 400 MW Chiba project.
But This Is Still a Plan
There is an important distinction.
The companies signed a:
Memorandum of Understanding, or MoU.
That means the $15 billion facility should not be described as if it already exists.
The Chiba project is planned.
Capital deployment across all phases is expected to exceed:
$15 billion.
Power capacity is planned at up to:
400 MW.
Operations are targeted to begin around:
2028.
Reuters reported that full capacity is expected in:
2029.
Large infrastructure projects can change.
Timelines can move.
Costs can rise or fall.
Plans can be revised.
The numbers describe the project currently being developed.
Not a finished data center.
The AI Bottleneck Can Move
For years, the AI race focused heavily on one question:
Who has the chips?
That made sense.
Advanced GPUs were scarce.
But imagine everyone eventually gets more GPUs.
Now they all need:
electricity,
data centers,
transformers,
cooling,
land,
fuel,
and financing.
Then the bottleneck can move.
The scarce thing may no longer be only the chip.
It may become:
the place where the chip can actually be turned on.
This is why power generators and data center developers suddenly matter so much to the AI story.
The chip may be the brain.
But the brain still needs a body.
AI Is Becoming a Physical Business
Software helped create enormous companies without requiring factories in every city.
AI looks similar on the screen.
You type a question.
A few seconds later, words appear.
It feels almost weightless.
But underneath those words can be:
power stations,
gas,
electrical cables,
transformers,
cooling systems,
industrial buildings,
and billions of dollars of computers.
The final product may appear on your phone as a few lines of text.
The infrastructure behind it is very physical.
The product may feel weightless.
The infrastructure is not.
So Why Build a $15 Billion AI Data Center Next to a Power Plant?
Because buying GPUs is not enough.
The GPUs have to turn on.
And hundreds of megawatts of electricity cannot always be delivered quickly.
Waiting years for a grid connection can mean waiting years to put billions of dollars of computing equipment to work.
That changes what makes a location valuable.
A power plant is no longer just somewhere electricity is produced.
The land beside it can become a shortcut to computing.
And in an industry where companies are spending billions to build AI infrastructure, that shortcut can be worth years.
The AI race may look like a race for better chips.
But another race is happening underneath it.
A race for:
the places where those chips can actually run.
If those places become scarce, something strange happens.
Land beside a power plant starts becoming part of the computer.
BEYOND THE OBVIOUS.