New Report Examines How AI Data Centers Are Increasing Electricity Costs

Across the county, utilities and public utility commissions are grappling with how to handle the billions of dollars of costs associated with providing electric service to proposed AI data centers. Utilities have long socialized the cost of infrastructure across their customer base. Now residential ratepayers and small businesses are facing rising costs, as electricity demand from AI data centers strains the grid and utilities race to build new gas power plants to serve AI data centers.

These costs include not just the cost of building new infrastructure such as power plants, substations, and transmission and distribution lines, but also operating costs, cost impacts of changes to congestion and outages, rising fuel prices, and the very real risk of stranded assets. The traditional approaches to cost allocation fail to address the unprecedented scale of the current buildout to meet AI’s energy demands, leaving other customers paying the price.

Earthjustice retained Energy Futures Group to examine the problems that new large-scale AI data centers pose to the traditional ways of allocating costs. The report Large Load Cost Allocation: Framework for States to Reduce Cross-Subsidization and Risk to Existing Customers includes potential solutions that would help ensure that household customers and small businesses do not end up footing the bill for AI data centers’ electricity.

The enormous scale of the loads proposed for AI data centers increases the threat of not only saddling other customers with their costs, but also constructing a grid that primarily benefits AI data centers at our expense, without improving reliability for the rest of us. For example, adding a data center on a transmission line that was originally built to bring power to a major city can totally change the flow of electricity on that line, creating new areas of congestion; the data center customer should pay for those new congestion costs, not everyone else.

The scale of energy demand from AI data centers is disrupting the electricity system in a number of ways that result in higher costs, such as increasing the price of fuels like natural gas — especially during times of peak demand — and distorting regional electricity markets, driving prices higher.

Utilities and state utility regulators must address the many ways AI data centers drive up costs for other customers:

• New power plants – Utilities are racing to build new power plants for AI data centers – most often gas generators – contributing to rising capital costs and equipment shortages.
• Interconnection and transmission and distribution costs – Most current U.S. grid systems do not have excess capacity to serve such large loads immediately.
• More expensive generation dispatch – Utilities are running higher-cost, less-efficient plants more frequently to meet the increased demand from AI data centers.
• Increased maintenance costs at power plants – Abrupt changes in electricity demand for AI data centers can result in more wear on generators as they cycle or ramp rapidly.
• Delayed power plant retirements – Older, less-efficient plants may be kept online to power AI data centers.
• Transmission and distribution congestion – Large loads such as AI data centers can create bottlenecks, preventing cheaper electricity from reaching customers.
• Network upgrades and maintenance costs – In order to maintain voltage, frequency, thermal limits, and reliability, additional upgrades to the grid may be necessary. Continuous use at higher levels accelerates wear on conductors and other transmission equipment.
• Forced-outage and system-response costs — Sudden loss of a very large data-center load can require emergency redispatch, additional fuel, voltage and frequency stabilization.
• Construction financing— Other customers may pay financing costs before the AI data center is operating.
• Stranded-asset risk — In the past, the risk of an industrial customer going out of business or failing to reach its projected electricity demand was manageable. But if a 500 MW AI data center ramps up slowly, significantly curtails its load, or ceases operations entirely, the costs incurred to support the data center will be shifted to the rest of the customer base.

Utilities can fail to explicitly consider these costs in rate cases because analyses of new large load customers are limited to either their incremental costs and benefits or specific cost categories. Continued use of existing cost allocation frameworks or customer class designations does not adequately address all the ways in which costs are being shifted from AI data centers to other customers. Traditional rates, riders, and allocators often spread the costs of serving AI data centers across other, existing customers instead of assigning them to the AI data centers that caused them.

The state utility commissions that regulate utilities’ rates for retail customers must revise their approaches to ensure that costs caused by serving AI data centers are paid for by those data centers.

For example, drawing on emerging practices in states such as Wisconsin and Oregon, the report assesses the possibility of directly assigning the costs of new electric generation facilities to the AI data centers that are creating the need for them. The paper also assesses best practices for ensuring that AI data centers are paying their fair share for the benefits they receive from the existing electric system.

The report includes additional suggestions for utility commissions such as:

• Minimum monthly bills and charges tied to contracted capacity rather than actual energy consumption.
• Collateral, exit fees, and enforceable term commitments to cover the risk that a data center delays, curtails, never reaches full load, or closes.
• More transparent interconnection, network upgrade, and other transmission cost allocation methodologies to allow the allocation of costs appropriately to new large loads.
• At the retail level, layered treatment of transmission and distribution costs: interconnection and network upgrades should be directly assigned to the new large loads requiring those upgrades, shared ongoing investments should be allocated via an updated cost of service study, and variable costs such as congestion should be handled using contribution-based factors.
• Hedging tools, such as Financial Transmission Rights, to offset the risk of rising congestion cost and assign costs while sharing any resulting revenues.
• Updated line extension policies should use a ‘but-for’ test that evaluates various factors to ensure that projects that will primarily serve a large load customer are paid for by that customer. Customer-specific transmission charges should be applied to recover costs once AI data centers are in-service.
• Revised construction financing agreements; update cost allocation and guardrails for Contribution in Aid of Construction (CIAC) agreements and allowance for funds used during construction recovery

Such guardrails, implemented with rigor by utilities and state utility regulators, can help ensure that AI data centers pay for the capacity, infrastructure, operating risks, and financial commitments they create rather than passing the cost on to other customers.

The research in the report builds on an earlier review of large load tariffs EFG completed on behalf of Earthjustice.

Earthjustice is working across the country to protect people from higher costs and increased pollution. Earthjustice attorneys appear before utility commissions in 22 states to ensure that AI data centers do not lock the U.S. into decades of expensive, polluting fossil fuels.