Who Pays for AI Power? Utilities and Regulators Are Starting to Decide
A quiet shift is underway in data center development.
For most of the past decade the conversation was straightforward. Communities competed to attract data centers. Developers evaluated land, tax incentives, and fiber connectivity. Utilities competed to offer power capacity.
The assumption underneath all of this activity was rarely questioned.
If new infrastructure was required to serve the load, the system would build it.
Now that assumption is being tested.
Across several states regulators are beginning to ask a different question. The issue is no longer whether AI infrastructure should be built. The issue is who pays for the electricity system required to run it.
That question is moving from the background of the industry into the center of development economics.
Cost Allocation Is Moving to the Forefront
AI-scale data centers have introduced a new kind of demand profile for power systems.
Traditional enterprise data centers often operate in the range of ten to twenty megawatts. A hyperscale AI facility can exceed one hundred megawatts. Some proposed campuses are far larger.
One proposed project in Delaware would require approximately 1,200 megawatts of electricity. That level of demand is roughly equivalent to the consumption of hundreds of thousands of homes.
When loads reach that scale, they do not simply plug into the existing grid. They often require new substations, transmission upgrades, and in some cases entirely new generation resources.
Historically those infrastructure costs were often absorbed into the broader utility system. Generation plants were built. Transmission lines expanded. The costs were recovered through rates paid by all customers.
The assumption was that growing demand would ultimately benefit the system.
That model is now under scrutiny.
Regulators are increasingly asking whether residential and small business customers should subsidize infrastructure built primarily for large AI data centers.
Early Signals from State Proceedings
The most visible example of this shift is unfolding in Delaware.
State regulators are reviewing a proposal from Delmarva Power to create a new large load tariff category for customers drawing twenty five megawatts or more. The proposed rule would define how infrastructure costs are allocated when massive electricity users connect to the grid.
The debate centers on a simple question. If new transmission lines or substations are required to serve a hyperscale facility, should those costs be paid by the developer or spread across the broader customer base.
Public comments in the proceeding show regulators probing deeper issues. Agencies are asking how much new demand utilities are forecasting, how congestion costs are being modeled, and what protections exist if a project scales back or fails.
The same theme is appearing in other states.
In California, a state oversight commission recently urged lawmakers to create new regulatory frameworks for large data center electricity demand. The report warns that billions of dollars in new grid infrastructure could be required to support AI growth and argues that the facilities driving those upgrades should bear the costs.
The commission recommended creating a special electricity rate category for very large power users and requiring financial commitments for the grid capacity they request.
Across these proceedings the language is remarkably consistent.
Data centers should pay the costs they impose on the power system.
Why This Matters for Site Selection
For developers and investors this shift carries real underwriting implications.
Most data center site selection models still prioritize land availability, tax incentives, and proximity to fiber networks. Power availability is considered, but the economics of that power are often assumed to remain stable.
Regulatory proceedings suggest that assumption is becoming fragile.
If utilities introduce new rate classes for large loads, the price of electricity can change after a project has already advanced. Transmission upgrades that might previously have been spread across the system could instead be assigned directly to the developer. Long term contracts or financial security requirements may be imposed before service is approved.
These decisions are not technical adjustments. They reshape project economics.
A campus that looks viable when evaluated on land price and available megawatts can become far more expensive if regulators require the developer to finance transmission expansion or commit to long term minimum demand charges.
What Developers Should Actually Be Evaluating
This is why the most important variable in data center site selection is no longer simply whether power exists.
The deeper question is how the regulatory framework assigns the costs of building that power.
Developers increasingly need to understand the history of rate design in a state. They need to examine whether regulators have approved separate tariffs for large loads. They need to study how transmission upgrades are allocated and how interconnection queues are prioritized.
The availability of megawatts matters. The structure of the electricity tariff may matter even more.
Utilities and regulators are beginning to treat hyperscale data centers as a distinct class of customer with unique impacts on the grid. Once that classification occurs, the rules governing cost recovery often change.
The New Constraint
The next phase of the data center market will not be determined by tax incentives or cheap land.
It will be shaped by how electricity is priced and allocated for AI scale demand.
Communities that once competed aggressively to attract hyperscale infrastructure are now confronting the reality that these facilities can reshape regional power systems.
The result is a new political economy around electricity.
The question is no longer simply whether AI infrastructure will be built.
The question regulators are increasingly asking is who will pay for the power system required to support it.