Every request for AI computing begins with electricity.

But the electricity is only part of the cost.

Delivering hundreds of megawatts to an AI campus may require new generation, transmission lines, transformers, substations, switchgear and distribution infrastructure. Utilities may need to plan and finance those systems years before the first server begins operating.

That creates a difficult question.

What happens if the project arrives late, uses less electricity than expected or never gets built?

The utility infrastructure may already be under construction. The debt may already have been issued. The equipment may have been ordered. Those costs do not disappear because the customer’s forecast changed.

Someone still has to pay.

Historically, utilities have recovered many infrastructure costs across broad groups of customers. That model becomes more complicated when one proposed facility requests as much electricity as a city and when several proposed facilities may be competing for the same capacity.

This is turning utility regulation into one of the most important policy arenas in the Intelligence Economy.

The debate is not simply about whether data centers use too much electricity.

It is about who carries the financial risk required to serve them.

The Short Answer

The cost of serving an AI data center can be divided among the data-center customer, the utility, utility shareholders and the broader body of ratepayers.

Who ultimately pays depends on utility tariffs, service agreements, regulatory decisions and the type of infrastructure being constructed.

Large-load tariffs are increasingly being designed to require major customers to make long-term financial commitments before utilities invest in new capacity.

Those Provisions Can Include:

• Load-study fees

• Construction deposits

• Minimum monthly payments

• Contracted-demand requirements

• Multiyear service terms

• Collateral or credit guarantees

• Load-ramp schedules

• Cancellation payments

• Exit fees

• Customer-funded infrastructure

• Curtailment or flexibility commitments

These mechanisms are intended to protect existing customers from paying for infrastructure created primarily to serve a project that does not materialize as expected.

But the tariff must also give the large customer a realistic path to power.

If the requirements are unclear, inconsistent or excessively restrictive, investment may move to another utility territory.

The goal is not merely to make the data center pay more.

It is to assign costs and risks to the parties best positioned to control them.

The Signal

Large-load tariffs are rapidly becoming a national infrastructure issue.

The Smart Electric Power Alliance’s Database of Emerging Large-Load Tariffs now tracks more than 100 proposed and approved utility tariffs and service rules across the United States.

The database shows utilities and regulators experimenting with customer commitments, cost protections, load thresholds, flexibility incentives and new procedures for connecting major loads. SEPA

Federal policy is moving in the same direction.

In March 2026, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed a federal Ratepayer Protection Pledge. According to the White House, the companies committed to build, bring or purchase the power needed for their data centers, cover required power-delivery upgrades and negotiate separate rates that remain payable even if the electricity is not fully used. White House

The pledge is an important policy signal.

The enforceable details, however, will still be shaped by utility contracts, state regulators and regional power-market rules.

A national commitment may establish the principle.

The tariff determines what happens when the bill arrives.

A Megawatt Is Also a Financial Obligation

A request for 500 megawatts may sound like a technical requirement.

To the utility, it is also a long-term financial forecast.

The utility must determine how much generation will be required, where the electricity will move, which substations need to be built and whether existing transmission can support the load.

Those decisions can trigger investments whose useful lives extend for decades.

The data-center customer may want power in three years. The utility may be financing infrastructure expected to operate for 30 or 40.

That mismatch creates risk.

The project may change because of technology, capital markets, power availability, permitting, customer demand or corporate strategy. A campus initially expected to consume 500 megawatts may ramp more slowly. It may stop at 250 megawatts. It may move to another market.

The utility cannot build half a transmission line after discovering that half the forecast disappeared.

This is why a large-load application is more than a service request.

It is a request for the electricity system to make a capital commitment.

How the Cost Reaches the Electric Bill

Utilities recover the cost of providing service through rates approved by public utility commissions or other governing authorities.

Those Rates Generally Recover Several Types of Costs:

• Generating or purchasing electricity

• Building and maintaining transmission

• Operating the local distribution system

• Financing infrastructure investments

• Maintaining reliability and reserve capacity

• Operating the utility

• Meeting regulatory and public-policy requirements

Some costs can be connected directly to one customer. A dedicated substation serving a single data-center campus is an obvious example.

Other investments may serve several customers or strengthen the larger grid.

That is where cost allocation becomes difficult.

A new transmission line might be required because of one large-load project while eventually benefiting an entire region. A generation facility may be built to support growing data-center demand but also provide broader reliability. A substation may initially serve one campus and later support additional development.

Regulators must decide how much of the cost was caused by the large customer, how much benefits the broader system and how the associated risk should be divided.

This principle is often described as cost causation.

Customers should generally pay costs they cause.

The difficult part is determining where causation ends and shared benefit begins.

The Risk Behind the Load Forecast

Utilities are receiving enormous numbers of large-load inquiries.

Not every inquiry will become an operating facility.

A developer may evaluate several locations before selecting one. Related companies may submit requests in multiple utility territories. Projects can also be announced before land, financing, permits or customers have been fully secured.

If every request is treated as certain, the utility may overbuild.

If utilities discount legitimate requests too aggressively, they may fail to construct infrastructure quickly enough.

This is the forecast problem.

A credible large-load process should help distinguish serious projects from speculative ones.

Utilities and Regulators May Examine:

• Whether the applicant controls the proposed property

• Whether financing has been committed

• Whether permits are progressing

• Whether equipment has been ordered

• Whether the project has identified end users

• Whether deposits have been paid

• Whether duplicate requests exist elsewhere

• How quickly the customer expects to ramp

• What happens if milestones are missed

This information does not guarantee the project will succeed.

It gives the utility a better basis for deciding when a customer inquiry should become a system investment.

When the Forecast Is Wrong

Consider a simplified example.

A data-center developer requests 400 megawatts. The utility begins constructing a substation and transmission upgrades based on that request.

The campus opens later than expected and initially uses only 100 megawatts. Several years pass before the load reaches 200 megawatts. The remaining capacity is never used.

The infrastructure was sized for the original commitment.

Without contractual protections, the unused portion can become a stranded investment—an asset whose cost remains even though the expected demand did not arrive.

Utilities may attempt to recover that cost through future rates. Regulators may disallow some spending. Shareholders may absorb part of the loss. Existing customers may see higher bills.

None of those outcomes is attractive.

That is why large-load tariffs increasingly require customers to pay for a minimum portion of the electricity capacity reserved for them.

The customer is not only buying electricity.

It is paying the system to remain ready to deliver it.

The New Large-Load Tariff

A utility tariff is a legally enforceable schedule of rates, charges and service conditions.

For ordinary customers, the tariff may feel almost invisible.

For an AI campus, it can shape the entire project.

Modern large-load tariffs may include several layers of protection.

1. Eligibility Threshold

The tariff defines which customers are considered large loads.

The threshold may be based on megawatts, load factor, projected energy use or a combination of measures.

This separates extraordinary projects from ordinary commercial and industrial customers.

2. Study Fees

Large-load studies require engineering, transmission analysis and system planning.

Utilities may charge customers to begin that work. A meaningful study fee can also discourage speculative applications submitted without serious project commitment.

3. Contracted Demand

The customer agrees to reserve a specified amount of capacity.

That commitment gives the utility a basis for designing and financing the infrastructure required to serve it.

4. Load-Ramp Requirements

Few data centers reach full demand on the first day.

A tariff may establish the percentage of contracted capacity the customer must reach during each year of development.

This creates a measurable path from initial energization to full operation.

5. Minimum Payments

The customer may be required to pay for a minimum percentage of contracted demand, even when actual use is lower.

This helps the utility recover the cost of infrastructure held available for the project.

6. Contract Duration

Large-load agreements may extend for a decade or longer.

Long terms align the customer’s financial commitment more closely with the infrastructure built to serve it.

7. Collateral

Customers may need to demonstrate credit quality or provide a financial guarantee.

Collateral protects the utility if the project sponsor cannot meet its obligations.

8. Exit Fees

A customer leaving early may be required to pay part of the remaining contract value or the unrecovered cost of dedicated infrastructure.

The purpose is to prevent an abandoned project from transferring its costs to everyone else.

9. Flexibility Provisions

Some tariffs may reward customers capable of reducing electricity use during periods of grid stress.

Flexibility can help the utility serve more demand without immediately constructing every piece of infrastructure required for fully firm service.

The tariff becomes more than a price.

It becomes the operating agreement between the AI campus and the electricity system.

Tariff Watch: AEP Ohio

AEP Ohio provides a useful example of how detailed these agreements are becoming.

Its data-center tariff applies to large requests above 25 megawatts and establishes study fees ranging from $10,000 to $100,000, depending on the amount of capacity requested.

Customers must provide a specific site, expected load ramp and final demand before the utility conducts its study.

The tariff allows a load-ramp period of up to four years. The initial contract term consists of that ramp period plus eight additional years.

It also includes minimum-demand charges, collateral requirements for customers that do not satisfy specified financial standards and financial responsibility when a customer cancels or materially delays a project before energization. AEP Ohio

The structure communicates something important.

A place in the power queue is becoming a financial position.

The customer cannot reserve hundreds of megawatts indefinitely without demonstrating commitment.

That may reduce speculative demand and give the utility greater confidence that infrastructure investments will be used.

It also changes the site-selection analysis.

The developer must evaluate more than the price per kilowatt-hour. It must consider study deposits, minimum payments, collateral, ramp requirements, contract length and the cost of changing its plans.

Protecting Ratepayers Without Blocking Investment

Ratepayer protection is necessary.

It can also be designed poorly.

A tariff that places every conceivable risk on the new customer may protect the utility but make the market commercially unattractive. Excessive collateral, rigid ramp schedules or punitive exit terms can make an otherwise strong site impossible to finance.

A tariff that is too permissive creates the opposite problem.

The utility may begin construction before the customer has demonstrated sufficient commitment. Existing customers can then inherit the financial consequences if the project fails.

The strongest policies find the balance.

They protect households and existing businesses while providing large customers with transparent terms, realistic milestones and a credible path to power.

That requires accountability on both sides.

Customers should not be allowed to reserve capacity without meaningful commitments.

Utilities should not be allowed to assign loosely related infrastructure costs to one customer simply because it has a large balance sheet.

Regulators must examine the relationship between the investment and the demand it is intended to serve.

Ratepayer protection is not achieved by automatically approving or rejecting infrastructure spending.

It is achieved by understanding who benefits, who controls the risk and who should be responsible when assumptions change.

Can Data Centers Lower Electricity Costs?

The rate discussion is often framed as though data centers can only increase costs.

The economics are more complicated.

Large data centers can consume electricity at relatively consistent levels throughout the day. When a high-load-factor customer uses existing infrastructure more fully and pays rates reflecting its cost of service, the additional revenue can spread fixed system costs across more electricity sales.

An Electric Power Research Institute analysis found that high-utilization AI data centers can theoretically reduce average system costs under the right conditions. But it also concluded that the benefit depends on rate design, infrastructure requirements, cost allocation and regulatory policy. EPRI

That qualification matters.

A data center using spare generation and transmission capacity is economically different from one requiring an entirely new power system.

A campus that reaches its forecast demand is different from one reserving capacity it never uses.

A customer paying the full incremental cost it creates is different from one receiving preferential terms recovered from other ratepayers.

Data centers do not automatically raise or lower rates.

The infrastructure required to serve them and the policy governing who pays for it determines the outcome.

Load Flexibility Becomes a Negotiating Asset

Most AI campuses want firm, uninterrupted electricity.

The grid increasingly values flexibility.

During a small number of high-stress hours, a data center capable of temporarily reducing grid demand could help avoid outages, reduce peak-capacity requirements or delay certain infrastructure upgrades.

Flexibility Might Come From:

• Shifting computing workloads to another region

• Rescheduling nonurgent training activity

• Temporarily reducing processor utilization

• Using batteries

• Operating approved onsite generation

• Adjusting cooling systems

• Participating in utility demand-response programs

Not every workload can move, and not every campus will be able to curtail safely.

But when flexibility is real, measurable and contractually enforceable, it can become a valuable infrastructure resource.

Federal regulators are examining how flexible large loads should be studied and connected to the transmission system. FERC’s large-load proceeding has considered whether customers agreeing to curtail during system stress could qualify for different or faster interconnection treatment. FERC

This could create a new exchange:

The customer provides operational flexibility.

The utility provides faster or less expensive access to power.

Flexibility is not a substitute for adequate infrastructure.

It can reduce the amount that must be available during the grid’s most constrained hours.

State Watch: Virginia

Virginia sits near the center of the large-load policy debate because of the scale of its data-center market.

In November 2025, the Virginia State Corporation Commission approved a new GS-5 rate class for Dominion Energy Virginia customers demanding at least 25 megawatts. The class is scheduled to take effect on January 1, 2027.

Certain large customers will be required to pay at least 85% of contracted distribution and transmission demand and 60% of generation demand, among other requirements. Virginia State Corporation Commission

The significance extends beyond one utility.

Virginia is demonstrating how a mature data-center market begins separating its largest customers into a distinct regulatory category.

The state is no longer treating a major data-center campus as simply another commercial electricity customer.

Its scale, load profile and infrastructure requirements justify different rules.

Other jurisdictions will be watching how those rules affect ratepayers, utility investment and future data-center development.

Utility Tariffs Are Site-Selection Policy

Economic-development teams often lead with taxes, land prices and incentives.

They should also understand the utility tariff.

Two properties with similar electricity prices may carry very different financial obligations.

One utility may require a larger deposit. Another may impose a longer contract. One may offer a defined ramp period. Another may require the customer to pay for most contracted demand almost immediately.

The Difference Can Affect:

• Project financing

• Construction phasing

• Customer commitments

• Site-control strategy

• Speed to power

• Operating costs

• Exit risk

• The value of behind-the-meter generation

• The ability to expand or transfer capacity

A tariff can make a market more credible by giving developers clear rules.

It can also make nominally available power difficult to use.

This is why utility regulation belongs inside the site-selection process from the beginning.

The cheapest electricity price does not always produce the lowest-risk power agreement.

The Large-Load Compact

The relationship between an AI campus and the electricity system should not be one-sided.

This is the Large-Load Compact:

Customer Commitment + Utility Accountability + Ratepayer Protection + Load Flexibility = Sustainable Infrastructure Growth

Each participant carries a responsibility.

The Customer

The customer provides credible demand forecasts, demonstrates site control, meets development milestones and makes financial commitments proportionate to the infrastructure being reserved.

If the project changes, the customer helps cover costs created by that decision.

The Utility

The utility provides transparent study processes, realistic energization schedules and clear explanations of the infrastructure required.

It distinguishes customer-specific investments from upgrades benefiting the broader system.

The Regulator

The regulator tests assumptions, evaluates cost causation, protects existing customers and ensures that utility shareholders and large customers carry an appropriate share of project risk.

The Electricity System

Where practical, the project provides measurable flexibility, onsite resources or other grid services that reduce peak demand and improve resilience.

The compact is not a single contract.

It is a principle for aligning the interests of the project, utility, investors and community.

AI infrastructure becomes more sustainable when the people creating the demand also participate in managing the system required to serve it.

Policy Across the Eight Pillars

Large-load policy touches every pillar of the Infrastructure of Intelligence™.

Power is the most direct connection. Tariffs determine how generation, transmission and substations are financed.

Connectivity and land are affected because utility terms can determine where a campus remains economically viable. Water and sustainability strategies can affect the facility’s electricity profile and operational flexibility.

Workforce demand grows when grid infrastructure expands. Capital responds to contract obligations, collateral requirements and the certainty of the energization schedule.

Policy connects all of them.

A tariff may look like a utility document.

In practice, it can influence site selection, project finance, campus design, construction phasing, sustainability strategy and community support.

The eight pillars remain separate areas of analysis.

The power agreement can change every one of them.

What Leaders Should Start Asking

Communities, regulators, utilities, developers and investors should move beyond asking how much electricity a data center needs.

They Should Ask:

• What infrastructure must be built to serve the load?

• Which investments serve only the data center?

• Which investments benefit the broader grid?

• Who pays the initial construction costs?

• How will those costs be recovered?

• What load has the customer contractually committed to use?

• How long is the load-ramp period?

• What minimum monthly payment is required?

• How long is the service agreement?

• What collateral or credit support must be provided?

• What happens if the project is delayed?

• What happens if the project is canceled?

• Are exit fees tied to actual unrecovered costs?

• Can reserved capacity be transferred to another customer?

• How does the utility identify duplicate or speculative requests?

• What assumptions are included in the utility’s load forecast?

• Can the customer reduce demand during system emergencies?

• How will behind-the-meter generation affect the agreement?

• Are existing customers protected from stranded costs?

• Can the utility demonstrate that its energization timeline is realistic?

• How does the tariff compare with those in competing markets?

These questions reveal the real economics beneath the megawatt announcement.

The Golden Nugget

The real policy question is not whether data centers use a lot of power. It is who pays when the forecast is wrong.

When the customer reaches full demand, a well-designed agreement can support infrastructure investment and potentially improve system utilization.

When the project falls short, the same agreement determines whether the risk remains with the customer or moves onto everyone else’s electric bill.

The tariff decides before the outcome is known.

The IOI Take

The utility tariff is becoming one of the hidden maps of the Intelligence Economy.

It influences where AI infrastructure can be financed, how quickly it can reach power and which regions can grow without transferring unreasonable risk to households and existing businesses.

The strongest markets will not promise unlimited electricity at any cost.

They will create transparent agreements balancing speed, reliability, affordability and customer responsibility.

They will require credible financial commitments before constructing infrastructure.

They will hold utilities accountable for realistic studies and schedules.

They will reward usable flexibility.

Most importantly, they will recognize that ratepayer protection and AI growth do not have to be opposing goals.

With the right structure, the companies creating the demand can help finance a stronger electricity system.

Without it, the AI buildout can leave communities carrying infrastructure risk they never agreed to assume.

One Last Thought

A megawatt is a measurement of power.

For an AI campus, it is also a promise.

It is a promise that the customer will arrive, the utility will build and the electricity system will be ready.

Policy determines who remains responsible when that promise changes.

Workforce

Policy

Capital

Sustainability