The $4.3 Billion Bill You're Already Paying for AI: How Utilities Shift Data-Center Costs to Ratepayers

Forbes asked the right question: when a utility builds substations, transmission lines, or power plants years before a data center is certain to arrive, who eats the cost if the project shrinks or stalls^[1]? The answer, visible now in rate filings and cost audits, is that residential customers are absorbing the gap between what utilities spend and what they recover from the hyperscaler.

A Union of Concerned Scientists analysis of just seven states within PJM's jurisdiction found $4.3 billion in data-center grid connection costs passed to consumers in 2024 alone^[9]. That bill covers infrastructure utilities built to serve large loads that may never materialize at full capacity, shrink, or delay indefinitely. The mechanism is brutally simple: utilities propose new generation and transmission in their integrated resource plans, anchor the capex request to signed (or merely announced) data-center pipelines, and the state regulator approves it as prudent investment. The utility recovers its return on equity by spreading the asset base across all ratepayers, not just the load that justified building it. If the data center delays three years or arrives at half the contracted power, the steel and copper are still in the ground, still earning a regulated return, still on your bill.

Researchers at Harvard Electricity Law Initiative have documented how this works: utilities secure special contracts with hyperscalers, often confidential, that lock in discounted rates or cost guarantees for the big customer while socializing the capex risk^[7]. The result is a transfer of profit from the public to the utility and the corporation. Wholesale electricity costs in areas near data centers have risen as much as 267% in five years^[5], and that wholesale pressure is passed through to retail customers who had no seat at the negotiating table.

The second distortion is the load forecast itself. Utilities now use data-center pipelines as rate-case weapons: inflated buildout projections justify capex programs that ratepayers fund even if the announced load never materializes. The forecast is negotiable, rarely audited by commissions, and often includes duplicate requests from developers shopping the same project to multiple utilities, phantom load that never appears on anyone's grid. A Goldman Sachs projection estimated AI infrastructure buildout would increase electricity costs by 6% between 2026 and 2027, and another 3% by 2028^[4]. Whether that cost lands on your bill or on the hyperscaler depends entirely on how the regulator split the contract terms and the asset assignment.

The protective tariff already exists in some states and can be demanded in any rate case: a large-load customer class with standardized rules that isolate risk from residential customers. The mechanism shifts the burden back to the load: high minimum take clauses (pay for 85% of transmission capacity whether or not you use it), long contract terms matched to asset life (10 to 15 years minimum, not five), collateral and exit fees covering unamortized investment, and 100% responsibility for dedicated network upgrades. Virginia, Ohio, and Oregon have versions on the books. The window to demand these terms in your state is now, in the next rate case or interconnection docket. Without them, the data-center boom funds itself on your bill.