Artificial intelligence may live in the cloud.

Its electricity bill does not.

Behind every AI model sits physical infrastructure:

• data centers,
• processors,
• cooling systems,
• power plants,
• substations,
• transmission lines,
• and an electrical grid capable of keeping the entire arrangement running.

As AI infrastructure expands, some data centers are becoming extraordinarily large electricity customers.

That creates an interesting problem.

A technology company can decide relatively quickly that it wants more computing capacity.

The electrical system cannot necessarily create the corresponding power capacity nearly as quickly.

And when AI needs more electricity than the grid can easily supply, somebody has to build more capacity.

Which leads to the more important question:

Who pays?

The Grid Wasn’t Built Around AI

The electrical grid is a shared system built over generations.

Homes use it.

Factories use it.

Hospitals use it.

Stores use it.

Schools use it.

Data centers use it.

Normally, demand changes gradually enough that utilities and grid operators can plan around it.

AI complicates that pattern because very large new loads can arrive faster than generation and transmission infrastructure can be built.

A data center therefore isn’t simply another customer plugging into an outlet.

At sufficient scale, the customer can change what the system itself requires.

A Private Electricity Need Can Create a Shared Infrastructure Need

Imagine a company wants to build a massive AI facility.

The facility requires far more electricity than the local system can currently deliver.

Several things might now be necessary:

• additional power generation,
• new transmission capacity,
• larger substations,
• distribution upgrades,
• new reliability measures.

The data center is privately owned.

The wider electrical system is shared.

That boundary is where the interesting economics begin.

If infrastructure must be expanded primarily because one enormous customer arrived, should that customer pay for all of it?

What if the upgrades also benefit other customers?

What if the company promises enormous electricity demand, infrastructure gets built, and the project later shrinks or disappears?

What if everybody benefits twenty years from now but one group has to finance construction today?

There is no magical accounting formula that makes these questions disappear.

Someone has to decide how the costs and risks are distributed.

Who Pays When AI Needs More Electricity Than the Grid Can Easily Supply?

There are several possible places for the cost to go.

The AI company can pay.

It can finance dedicated infrastructure, pay connection and upgrade costs, contract for new generation, or accept electricity rates designed around the costs it creates.

The utility can invest.

A utility can build infrastructure and recover its costs through rates over time.

Government can participate.

Public financing, incentives, infrastructure programs, favorable tax treatment, or other development policies can absorb portions of the cost.

Existing customers can potentially absorb costs.

Depending on the rules, some costs of expanding a shared electrical system can eventually appear in the rates paid by households and businesses.

And there is another possibility:

the cost can be divided among several of them.

This is why “Who pays?” is more useful than “Who built it?”

The Stranded-Cost Problem

There is another layer that receives less attention.

Suppose a utility expects enormous AI demand.

It builds infrastructure around that expectation.

Then something changes.

The technology becomes more efficient.

The company relocates.

The facility is delayed.

The expected demand never fully materializes.

The infrastructure still exists.

And somebody still financed it.

This is known as stranded-asset risk.

It exposes one of the most important questions in infrastructure economics:

Who carries the risk when private projections create public or shared investment?

The answer matters as much as the original construction cost.

Why Regulators Are Paying Attention

This is not merely theoretical.

The rapid growth of data centers has forced utilities and regulators to reconsider how very large electricity customers connect to and pay for the grid.

One objective is straightforward:

Make sure the arrival of an unusually large customer does not automatically shift the costs it creates onto customers who had nothing to do with the decision.

That can mean special rate structures, minimum payment commitments, cost-recovery agreements, requirements to finance upgrades, or obligations to bring additional power supply.

In other words, the system can be designed to resist cost externalization.

That point matters.

Cost Shifting Is Not a Law of Nature

Vampire Guides examines how costs tend to migrate toward people with less leverage.

But “tend to” matters.

Architecture can change the outcome.

If rules require a large electricity customer to pay the incremental costs it creates, the cost stays closer to its source.

If those protections are weak, ambiguous, politically negotiable, or based on overly optimistic assumptions, more risk can migrate outward.

The important variable is not corporate virtue.

It is system design.

Why the Largest Customer Has Leverage

There is also a negotiation problem.

A household requesting electrical service has very little bargaining power.

It needs electricity considerably more than the utility needs that particular household.

A multibillion-dollar development is different.

Large projects can bring:

• investment,
• tax revenue,
• construction activity,
• employment,
• political prestige,
• future economic development.

Regions may compete to attract them.

That gives a sufficiently large customer something ordinary customers rarely possess:

the ability to negotiate around the system rather than merely accept its terms.

And that leads directly into another VG mechanism:

Size does not merely increase consumption.

Size can increase leverage.

The Five-Layer Map

The functional hierarchy helps clarify what is happening:

Deciders → Creators → Operators → Enforcers → Everyone Else

• Deciders determine how aggressively AI development and new electricity capacity should be pursued.
• Creators design rates, contracts, infrastructure plans, incentives, and cost-allocation rules.
• Operators build and run the electrical system.
• Enforcers administer the resulting requirements.
• Everyone Else eventually experiences the resulting reliability, prices, benefits, and risks.

The important decisions therefore happen long before an ordinary customer opens an electricity bill.

By then, the architecture has largely been established.

AI Isn’t Really the Point

AI makes this mechanism unusually visible.

But the mechanism is older than AI.

A factory can create infrastructure requirements.

A housing boom can create them.

A new industrial district can create them.

A sports stadium can create them.

Any sufficiently large private project can eventually encounter the limits of shared capacity.

At that point, growth becomes a distribution question.

Not simply:

Should we build this?

But:

Who receives the benefits?

Who finances the enabling infrastructure?

Who carries the downside risk?

The Clarifying Insight

Who pays when AI needs more electricity than the grid can easily supply?

There is no automatic answer.

And that is the point.

Costs do not distribute themselves.

Rules distribute them.

Contracts distribute them.

Rate structures distribute them.

Political decisions distribute them.

Bargaining power influences where they finally land.

So when someone announces billions of dollars in exciting new investment, don’t stop at the size of the investment.

Look underneath it.

Ask what new infrastructure becomes necessary.

Then ask three questions:

Who benefits?

Who pays?

Who carries the risk if the assumptions are wrong?

That’s where the architecture becomes visible.

Want the larger map?

The Vampire Playbook explains how modern systems distribute power, costs, risks, and consequences—and why they so often travel in different directions.

Get the Vampire Playbook