Who Pays to Power AI Data Centers? Comparing BYOG and Backstop Capacity Procurement

Key Takeaways

• AI-driven data centers are creating a spike in electricity demand and necessitating costly investments that new and/or existing utility customers must pay.
• State and federal policymakers are in a bind: data centers are a strategic imperative, but consumer backlash is becoming a major political concern and leading to proposed moratoriums on new buildouts.
• Two frameworks are emerging to navigate these challenges: co-location/bring your own generation (BYOG) and backup capacity procurement. Each has its own opportunities, risks, and outstanding questions.

Artificial intelligence (AI) has upended the once‑steady trajectory of US electricity demand. After decades of essentially flat growth, planners and regulators face a surge driven by hyperscale data centers—massive campuses with energy needs measured in the hundreds of megawatts—straining long‑standing planning assumptions and accelerating calls for new infrastructure.

This sea change collides with a grid built for slow, incremental expansion. Generator interconnection queues stretch for years—up to seven in the heart of “Data Center Alley” in Virginia, the world’s largest hub of data centers. Transmission projects must navigate protracted permitting and construction timelines. Existing capacity market designs depend on predictable, gradual increases in demand to produce prices that are reliable for suppliers and reasonable for customers.

Put simply, demand for power is outpacing supply and deliverability at a speed the grid has never experienced. As a result, utilities are paying more to procure and deliver energy—costs that ultimately flow through to customers. This creates a difficult balancing act for federal and state policymakers. On one hand, rapidly expanding AI infrastructure is a strategic imperative. On the other, soaring electricity prices are becoming a political liability, fueling consumer frustration and prompting calls for moratoriums on new data centers.

In this article, we compare two emerging frameworks to navigate these tensions: co‑location/bring your own generation (BYOG) models and new forms of supplemental or “backstop” capacity procurement.

Who Pays for Growth? Costs, Risks, and of Two Emerging Paradigms

Market participants, customers, and policymakers face two fundamental categories of risk as they grapple with who should pay how much for which components of the grid—generation, transmission, and distribution—and on what basis:

1. Overbuilding and underbuilding: Large‑scale energy infrastructure takes years to plan and construct, so supply and demand are impossible to perfectly synchronize. Overbuilding occurs when utilities invest in new infrastructure to serve data center demand that ultimately does not materialize. Underbuilding occurs when utilities do not make sufficient upgrades to serve both existing customers and rapidly expanding, AI-driven demand. An overbuilt system is more robust and expensive than necessary to maintain full reliability. An underbuilt system, by contrast, cannot guarantee that all users will have access to the full amount of power they are willing to purchase.
2. Cost allocation: Given that grid costs have historically been shared across all customer classes, non–data center customers may end up funding infrastructure built to meet the data center demand surge—even in the absence of overbuilding. This raises long‑standing questions of equity, fairness, and regulatory responsibility that are taking on new urgency.

These costs and risks have always been fundamental to utility regulation but are especially thorny today because a relatively small set of new customers is driving exceptionally large incremental expansion needs.

Two major frameworks for managing these risks have emerged.

Framework 1: Co-location/Bring Your Own Generation (BYOG)

What BYOG Is

Traditionally, utilities transmit and deliver electricity from centralized generators to customers dispersed across the grid. Co‑location turns this model on its head: customers site their energy needs directly beside new or existing generation, eliminating the need for additional or upgraded delivery infrastructure.

In this arrangement, the data center itself typically finances new capacity through bilateral contracts. BYOG projects still often rely on the grid for backup power and restart capabilities. Therefore, the Federal Energy Regulatory Commission (FERC) favors co-located customers paying for the transmission, distribution, and grid services they consume.

Where It’s Already Happening

One prominent example is Microsoft’s twenty year power purchase agreement for the output of Constellation Energy’s Three Mile Island nuclear plant. Last November, the Department of Energy (DOE) backed the nuclear restart with a $1 billion loan, in large part to help make electricity cheaper for consumers on the associated power grid.

Risks and Cost Allocation Questions

• For all customers: If co-located generators fail and data centers—which maintain interconnections for backup power—pivot to grid supply, the result can be stress on grid reliability.
• For generators: If the data center exits or fails, the generator may have no easy alternative buyer without transmission deliverability.
• For non–data center customers: What happens if data centers don’t pay for the grid services they use? FERC is trying to mitigate these risks by, for instance, ordering PJM to establish clear, nondiscriminatory interconnection rules for generator-connected loads and offer new transmission services to prevent cost-shifting.

Strategic Benefits and Market Opportunities

• For data center customers: Bypassing system planning and major transmission expansions can bring supply online faster. Customers can also choose clean or renewable sources.
• For non–data center customers: They avoid capacity cost hikes when data center customers bring their own power.
• For generators/developers: Long-term bilateral contracts can provide more stable revenue than annual capacity auctions. They also could mitigate counterparty risk by proactively seeking locations more favorable for future interconnection or industrial non–data center customers.
• For energy storage/distributed energy resources (DERs): Energy storage providers have an opportunity to contract data centers to provide grid capacity when local generation is out, while nearby DERs could provide energy directly to the data center itself. However, these opportunities require the existence of applicable tariffs and DER accreditation schemes, which remain to be written in several areas.
• For utilities/transmission owners: Should co-located data centers cease operations, future transmission development opportunities will arise linking stranded generation up to the grid—though significant cost allocation debates will likely occur.

Framework 2: Backstop or Supplemental Capacity Procurement

What Backstop Procurement Is

Backstop procurement occurs when regional transmission organizations (RTOs), independent system operators (ISOs), or adjacent procurement authorities obtain generation capacity outside existing mechanisms. The objective is to avert grid reliability risks that standard auctions and development pipelines cannot meet in time.

Under this framework, transmission upgrades—with long lead times and high capital expenditures—are often unavoidable. How those costs are allocated across customer classes remains unsettled. FERC lacks jurisdiction over retail rates and cannot shield all customers from potentially absorbing these incremental expenses. Ultimately, state utility commissions will decide how these costs are assigned.

Where It’s Already Happening

Earlier this year, PJM—after its last capacity auction hit record prices but fell short of its reliability target by 6.6 gigawatts—announced a “reliability backstop” capacity procurement. While details remain to be worked out, the key contention will undoubtedly be who pays. In a “Statement of Principles Regarding PJM,” state and federal officials suggested that costs should be allocated to data centers that have not obtained BYOG power.

Cost Allocation and Overbuild Risks

• For all customers: If even a portion of AI load is transient, rapid backstop procurement results in significant overbuilding. Accurately forecasting load growth is crucial in this approach.
• For non–data center customers: If backstop resources are obtained via long-term contracts, as opposed to existing annual auction constructs, payment obligations likely will shift to remaining customers in the event of data center exits.
• For existing generators: Backstop capacity that later enters standard energy and capacity auctions could depress prices and trigger retirements, potentially chilling future investments should existing market participants see backstop procurement as unpredictable competition.
• For utilities: If utilities undertake major transmission projects to serve data centers that never materialize, they and/or their customers will have to cover the costs. This has given rise to transmission security agreements (TSAs) between data centers and utilities, wherein data centers promise to make certain minimum payments to cover transmission upgrades for a given length of time whether or not they use service during that period. Yet TSAs alone do not shield non–data center customers from higher costs; states still must allocate incremental costs between customer classes.

Strategic Benefits and Market Opportunities

• For all participants: Aligning supply and demand over time through an integrated transmission-planning process lessens the chance of stranding new generation. Unlike with BYOG, backstop resources are typically designed with broader deliverability in mind.
• For RTOs/ISOs: Aside from backstop procurement, additional opportunities exist to revise standard auction parameters (e.g., by lengthening commitment terms) to incentivize participation and reduce the need for future backstops.
• For generators: Market participants may benefit from anticipating and positioning for capacity market reforms and how backstop resources will be treated in standard auctions if demand falters.
• For utilities/transmission owners: They will build substantial new transmission with a higher rate base leading to higher revenues.
• For energy storage/DERs: These players can go beyond capacity market participation into congestion relief. However, mechanisms for measuring avoided costs and necessary tariff structures remain to be implemented in many areas.

Enduring debates

The underlying question of who pays for expansion is anything but new. FERC and state public utility commissions have long grappled with complex cost-allocation challenges. Guiding principles—such as cost causation (the customer who causes a cost ought to pay for it) and beneficiary pays (a customer who benefits from a project should pitch in for it)—are well established. These foundations anchor today’s debates.

Both approaches can align with longstanding regulatory touchstones in their own ways.

Under BYOG, new data centers directly finance generation and aim to avoid triggering major transmission upgrades. Regulators still retain authority to ensure these facilities pay for specific grid services they do, in fact, use.

Under a backstop procurement model, regulators confront a broader allocation task: capacity and transmission investments, along with associated grid services, must be assigned thoughtfully across customer classes. Yet backstop capacity is simply additional system capacity, a category of costs that regulators have been allocating for generations.

Neither path offers a silver bullet. New firm generation cannot be deployed overnight, regardless of regulatory structure. Market forces—not regulators—determine the price of natural gas, turbines, and other critical equipment. Meeting rising data center demand will require pursuing additional strategies in parallel, including greater reliance on distributed energy resources and storage. Both technical opportunities to strengthen the grid and the long‑standing debate over how to allocate costs will persist even after the immediate surge in demand is addressed.