AI data centers push U.S. electricity costs higher, Goldman projects

Goldman Sachs analysts find that a surge in large, GPU-dense data centers focused on advanced AI workloads has outpaced new generation and grid upgrades, lifting wholesale and retail electricity prices across multiple U.S. markets. Measured year-over-year, electricity costs jumped sharply in 2025 and Goldman expects another sizable rise through 2027 before inflationary pressures ease in 2028 as fuel costs moderate. The incremental investment and operating costs utilities face to secure reliable supply — from new plants to transmission and substations — are largely being recovered through higher bills, a transfer that falls disproportionately on households with lower incomes and on regions hosting concentrated data center activity. Markets with clustered compute hubs and tight capacity procurement, notably the PJM interconnection, have seen particularly large system-cost increases; Goldman attributes multibillion-dollar liabilities in those territories to the load growth from hyperscale computing. At the same time, local opposition and permitting hurdles have reshaped or paused projects in states including Texas, Georgia, Illinois and Mississippi, industry trackers estimate roughly $64 billion of planned U.S. data-center projects are delayed or canceled — a dynamic that raises financing, execution and underutilization risks for developers and lenders. Financing patterns are evolving (corporate bonds, CMBS, syndicated loans and bespoke structures) and increasingly concentrate exposure to a small set of hyperscalers that anchor demand, making underwriting more sensitive to schedule and permitting risk. The gap between capacity under construction and verified, steady-state workloads also risks extended periods of low utilization, compressing returns on capital-intensive campuses built for future peaks rather than continuous load. Policymakers and regulators are responding: federal and state officials are negotiating contribution arrangements that could require large consumers to help fund generation or grid upgrades, while some cloud providers have offered voluntary limits or mitigation commitments to blunt price impacts. Technical and commercial mitigation tools — workload scheduling, demand-response agreements, on-site generation and storage, bespoke connection agreements and developer-funded substations — can reduce peak-driven capital spending but add complexity and cost. International contrasts sharpen the policy choices: some countries, notably China, are fast-tracking coordinated generation and transmission to enable rapid cluster commissioning, which can lower marginal power costs but also risks creating capacity mismatches and different environmental trade-offs. The short-term picture is constrained by limited labor and materials for new builds, keeping wholesale prices elevated in key states; longer-term relief will depend on a mix of additional firm generation, faster transmission and distribution investment, demand-side measures, and clearer cost-sharing frameworks. For utilities, regulators and investors the immediate challenge is balancing reliability, affordability and equitable allocation of costs as energy-intensive digital infrastructure scales, while political salience around rising utility bills is already influencing electoral contests and regulatory postures. Investors and corporate strategists should expect sustained capital spending on grid resilience, shifts in underwriting and site-selection criteria, and a likely period of consolidation if construction outpaces validated, steady workloads.