Who Pays When America’s AI Power Demand Doesn’t Show Up?

A company can ask for hundreds of megawatts of electricity before its data center is fully built.

The grid cannot make the same promise in reverse.

It has to plan substations, transmission lines, transformers and other infrastructure years before a new facility reaches full power.

That creates a financial problem hiding inside America’s AI boom: A power request is not the same thing as a power obligation.

And the difference is becoming expensive.

As of August 3, 2026, ERCOT was considering more than 474 gigawatts of requests to connect to the Texas grid — more than five times the state’s record peak demand. About 90% of those requests were associated with data centers. On August 3, Governor Greg Abbott ordered ERCOT and the Public Utility Commission of Texas to verify projects moving through the process before they could advance.¹

Texas large-load requests

474 GW+

ERCOT was considering more than 474 GW of requests to connect to the Texas grid as of August 3, 2026. This is a request volume, not a forecast of operating data-center demand.

That does not mean 474 GW of data centers are about to be built.

It means the electric system is confronting a harder question: Which future demand is real enough to become a financial and infrastructure commitment?

And increasingly, the answer is being determined with money.

A Request Is Not a Load

Several very different numbers can describe the same proposed data center.

There is the amount a developer requests.

Then the amount that satisfies the requirements to enter an interconnection study.

Then the amount the grid can actually serve.

Then the amount backed by a contractual or financial obligation.

Finally, there is the electricity the facility actually consumes.

Those are not the same thing.

ERCOT’s large-load process increasingly reflects that distinction. Its published materials describe development, technical and financial conditions that qualifying projects must satisfy as they advance through the interconnection process. The framework is designed to make large-load requests more than simple claims on scarce grid capacity.²

So Texas is no longer asking only:

“How much electricity might these companies eventually need?”

It is asking:

“Which requests are developed and committed enough to deserve scarce grid capacity?”

That distinction is the first financial lesson of the AI power boom:

Key finding

The largest number in the queue is not necessarily the number the grid should build around.

Texas Is Putting a Financial Gate in Front of the Grid

ERCOT’s Batch Zero framework attaches financial consequences to large-load requests.

Its published materials describe $50,000 per MW in financial security at the relevant large-load stage, along with site-control requirements and additional financial commitments as projects move toward interconnection. At the interconnection-agreement stage, the framework also includes a nonrefundable interconnection fee and consequences for withdrawal.²

Illustrative financial security

$50 million

For a 1,000-MW project, $50,000 per MW represents $50 million of financial security. This is a calculation based on ERCOT’s published $50,000/MW framework.

That does not mean the developer has paid $50 million to buy electricity.

It means the grid is requiring the developer to put meaningful money behind its request.

The question changes from:

“How much power do you want?”

to:

“How much are you willing to stand behind?”

That is a major shift in how the grid treats future AI demand.

Then Texas Added Verification

The timing matters.

ERCOT had been working toward conditional classifications for Batch Zero, but on August 31 it announced that it was delaying their release because additional data validation and due diligence were still required. ERCOT said the classifications would be released later in the week.³

That followed Abbott’s August 3 directive for a comprehensive audit of data centers seeking to advance through ERCOT’s interconnection process.¹

The sequence is revealing.

Texas had already created technical and financial gates.

Then it added another question: Is the project itself sufficiently real and verified to advance?

The size of the queue makes that distinction important. The state’s audit directive specifically requires information on each project’s projected electricity consumption, financing and financial assistance, ownership, on-site generation and other factors before projects move forward.¹

This does not prove the AI power boom is imaginary.

It proves something more useful: A developer’s requested megawatts are no longer enough, by themselves, to establish credible future load.

Ohio Shows What Happens When the Request Becomes a Contract

Texas is building financial discipline into the interconnection process.

Ohio shows what happens when the relationship gets closer to an actual service commitment.

In July 2025, the Public Utilities Commission of Ohio approved AEP Ohio’s enhanced requirements for large new data-center customers.

The key provision is straightforward: A large new data center must pay for at least 85% of the energy it has contracted for, even if it uses less.

The framework also requires financial-viability evidence and includes an exit fee if a project is canceled or fails to meet its obligations. The requirements operate over a 12-year period, including a four-year ramp-up.⁴

The economic principle is simple.

The customer is no longer merely saying:

“I might need this much electricity.”

The utility is saying:

“If you reserve this much capacity, you must accept financial responsibility for a substantial portion of it.”

That protects other customers from one obvious risk: infrastructure being built around a large projected load that later uses substantially less.

But it does not eliminate every risk.

The Contract Can Be Shorter Than the Infrastructure

This is where the story becomes much bigger than Texas.

A data center can sign a contract.

The utility can obtain collateral.

The customer can agree to minimum payments.

But the transmission system built around that customer can last for decades.

A Federal Energy Regulatory Commission concurrence makes the mismatch unusually clear.

In a February 2026 concurrence concerning a ComEd transmission security agreement, FERC Commissioner Judy Chang examined a hypothetical 600-MW customer that agreed to pay for at least 75% of its anticipated transmission service.⁵

That creates a minimum commitment equivalent to 450 MW.

But Chang’s analysis showed what happens if that customer triggers more than roughly $200 million in transmission upgrades.

The customer’s contractual payments can provide protection while the broader transmission costs may still be reflected in rates paid by other customers if the upgrades are sufficiently large.⁵

And then comes the critical mismatch:

The residual-risk problem

The agreement may last roughly 8–10 years. The transmission assets can have a useful life of roughly 40 years.

So the customer’s financial obligation can end long before the infrastructure does.

The wires remain.

The substations remain.

The capital investment remains.

And the costs associated with those assets can continue through the transmission rate structure.

A contract can protect customers against part of the risk without eliminating the infrastructure risk itself.

The Money Trail

The financial machinery emerging around large AI loads can be reduced to a simple chain:

Stage	Financial question
Power request	How much power does the developer say it may need?
Project verification	Is the project developed and credible enough to advance?
Interconnection eligibility	Can the grid accommodate the proposed load?
Financial security	How much money is the developer required to put behind the request?
Service or transmission commitment	What costs does the customer remain responsible for?
Grid infrastructure	What gets built before the facility reaches full operation?
Actual load	How much electricity does the facility ultimately consume?
Residual risk	Who carries costs that remain after the customer’s obligation ends?

Every step changes the question.

At the beginning: How much power does the developer want?

Then: How much can the grid reliably provide?

Then: How much money is the developer willing or required to put behind the request?

Finally: If the infrastructure exists but the expected load does not fully materialize, who carries the remaining cost?

That last question is the one that turns an electricity story into a financial story.

The Protection Is Real. So Is the Residual Risk.

It would be wrong to conclude that utilities are simply building billions of dollars of infrastructure for imaginary AI projects.

There are real data centers.

There are real customers signing agreements.

There are real financial commitments.

And there are real physical constraints requiring new transmission and distribution investment.

Exelon said in August that its transmission security agreements had secured more than $1 billion in customer protections by requiring large new power users to commit to projected transmission-service costs.⁶

Those protections matter.

So does what they reveal.

Utilities and regulators are redesigning the financial rules around large loads so that demand claims carry consequences before infrastructure is committed.

FERC is doing the same thing at the national transmission level.

In June, FERC issued show-cause orders to all six regional transmission organizations and independent system operators under its jurisdiction, directing them to justify or reform tariffs governing data centers and other large energy users. The Commission tied the action to faster large-load integration and stronger consumer safeguards.⁷

But “protected” does not mean “risk-free.”

Chang’s analysis shows why.

A customer commitment can cover a defined revenue obligation while leaving the broader transmission system exposed if the infrastructure required is unusually large relative to the committed load.⁵

So the regulatory response is not:

“AI demand is fake.”

It is:

“AI demand is large enough, uncertain enough and financially consequential enough that the rules have to change.”

Three Numbers America Should Stop Treating as One

The AI electricity debate often collapses several fundamentally different quantities into a single headline number.

Number	What it means	Why it matters
Requested power	What developers say they may eventually need.	The most speculative measure.
Financially committed power	The portion backed by contracts, collateral, minimum-payment obligations or other enforceable commitments.	More meaningful to the utility and its financial planning.
Operating power	What the facility actually consumes.	The quantity that ultimately matters to the physical grid.

Confusing these numbers creates bad conclusions in both directions.

Treat every request as certain demand, and you can overstate how much infrastructure must be built.

Treat every request as imaginary, and you miss the enormous amount of real investment already underway.

The important territory is between those extremes.

That is where the financial system is being rebuilt.

America Is Turning Electricity Requests Into Financial Tests

The biggest change may not be the amount of electricity AI will eventually consume.

It may be the way the grid decides which future consumption deserves scarce infrastructure.

Texas is requiring large-load projects to satisfy increasingly explicit development, technical and financial conditions.

Ohio is requiring large data-center customers to accept minimum-payment obligations and demonstrate financial viability.

Utilities such as Exelon are using transmission security agreements to obtain financial commitments from large new users.

And FERC is examining whether those arrangements adequately protect other customers when large loads require major transmission investment.⁶⁷

That creates a new hierarchy of credibility:

A request is a signal.

A study-qualified project is stronger.

A financially committed project is stronger still.

An operating facility is the final proof.

The gap between those stages is where the financial risk lives.

The Risk Has Not Disappeared. It Has Been Assigned.

America is not simply trying to produce enough electricity for artificial intelligence.

It is trying to determine who should financially stand behind the infrastructure required to produce it.

The new rules are an attempt to move that responsibility closer to the companies creating the demand.

But the FERC example shows the limit.

A customer can provide collateral.

A customer can sign a long-term agreement.

A customer can agree to pay a minimum amount.

None of those automatically makes the underlying transmission infrastructure disappear if the customer later scales back.

The infrastructure can remain.

The financing can remain.

The rate-base consequences can remain.

And the economic consequences can outlive the original contract.

There is no established public case in the evidence examined here showing that a canceled AI data center has already left ordinary customers with a quantified, realized loss of this exact type.

But the mechanism for residual risk is documented.

That distinction matters.

Because the real question is no longer:

“Will AI need enormous amounts of electricity?”

It probably will.

The more consequential question is:

How much of the electricity America is planning for will become a firm obligation — and who pays for the infrastructure built around the part that doesn’t?

That is the financial story hiding inside America’s AI power boom.

Sources

1. Office of the Texas Governor — Governor Abbott Directs Comprehensive Data Center Audit, August 3, 2026. Source
2. ERCOT — Large Load Update / 2026 interconnection gating framework, including site control and financial-security requirements. Source
3. ERCOT — Market Notice M-A080326-03, Delayed Issuance of Batch Zero Conditional Classifications, August 31, 2026. Source
4. American Electric Power — AEP Ohio Proposal on Data Centers to Protect Ohio Consumers Adopted by PUCO, July 9, 2025. Source
5. Federal Energy Regulatory Commission — Commissioner Chang’s Concurrence to Transmission Security Agreement between ComEd and Karis Critical, February 17, 2026. Source
6. Exelon — Exelon Secures More Than $1 Billion in Customer Protections Through Pioneering Transmission Security Agreements, August 4, 2026. Source
7. Federal Energy Regulatory Commission — FERC Launches Aggressive Targeted Action to Speed Large Load Integration, June 18, 2026. Source

Written and edited by Hossam Seif, founder of Money Traces.