Data centre construction activity across the United States has expanded at a pace not previously recorded. Fuelled by growing demand for artificial intelligence training and inference workloads, construction has been expanding across multiple states, from Virginia to Texas.

Power infrastructure has struggled to keep up with growing consumption. Power providers face operational pressure, grid managers issue advisories, and expenses rise across the board.

On Tuesday, Moody's Ratings released its formal evaluation, identifying electricity supply constraints as the leading obstacle holding back growth in the data centre sector. New development is outpacing the power grid's capacity to supply electricity.

Projections cited in the Moody's analysis estimate that by 2030, server facilities may draw as much as 426 terawatt-hours of electricity per year. According to International Energy Agency data referenced in the report, that figure stands at roughly twice the 224 TWh forecast for 2025. Fulfilling that increase would call for approximately 45 gigawatts of additional generation capacity, at an estimated cost of around $110 billion.

Annual costs across the entire electricity network would rise by between $25 billion and $30 billion, according to the analysis. Planned grid expansions that have received approval or are currently in progress are projected to cost between $80 billion and $115 billion by 2030, with that figure potentially growing threefold over the following ten years.

Extended equipment delivery timelines, approval processes that can take years, and multiple layers of compliance requirements all add to the difficulty. Conventional power infrastructure faces mounting difficulty keeping pace with rapidly expanding electricity demand, particularly in areas with high concentrations of data centres.

Regions such as Northern Virginia, Phoenix, and areas of Texas and Louisiana draw development interest through affordable land, tax incentives, and established fibre infrastructure. These locations also fall within areas where both power generation and grid transmission infrastructure are already operating near capacity.

A McKinsey analysis of power grid conditions across individual states identified approximately 40 GW of available dispatchable capacity for near-term expansion, along with 100 GW of new generation already under commitment. Between 50 and 75 GW of coal and gas generation capacity is expected to be decommissioned, while anticipated demand increases are projected to reach approximately 120 GW. The analysis projects a nationwide capacity gap of between 30 and 55 GW by 2030.

Spanning much of the Mid-Atlantic and Midwest, PJM Interconnection operates as the United States' largest electricity grid manager. The latest capacity auctions fell short of established reliability benchmarks, and projections for peak electricity demand continue to climb, driven largely by the expansion of data centres.

Research commissioned by Pennsylvania's electricity oversight authority projected that power generation capacity could fall short of consumer demand approximately six times per decade by 2030, under baseline forecast assumptions. Under conditions of elevated demand and constrained supply, that figure rises to 13 days annually. Projections from the same research indicate that consumption across the area could approach twice its current level by 2040, with data centres accounting for the vast majority of that growth.

Governor Greg Abbott directed a review of the state's extensive grid interconnection queue earlier this month. Of the 474 GW in total applications, data centres account for around ninety percent, while the freeze puts roughly 50 GW of pipeline capacity on hold. According to BloombergNEF estimates, that figure accounts for roughly one-fifth of all data centre capacity currently planned across the United States.

Projected financial shortfalls tied to postponed infrastructure capacity — with AI workloads accounting for 60 percent of the delay — may approach $8 billion by early 2027. A Jefferies analysis from June indicated that approximately half of the new U.S. data centre capacity planned for 2026 was at risk of missing its projected delivery timeline.

Grid capacity limitations contribute alongside land-use regulations, approval processes, and interconnection backlogs. Analyses from financial institutions indicate that between 30 and 50 percent of planned ventures encounter delays or are abandoned before completion.

One technology company aims to grow its data centre portfolio to 38 GW by 2032, a figure that would represent more than three times its current footprint. Contracts for AI and cloud services have gone unfulfilled owing to insufficient supply capacity. Some companies have opted for on-site power generation through gas turbines, compact nuclear reactors, and battery storage combined with renewable energy sources.

According to SemiAnalysis data, confirmed orders for dedicated AI computing power infrastructure have reached 75 gigawatts. On Tuesday, the Federal Energy Regulatory Commission directed six regional grid operators to either defend or revise their tariff structures governing high-volume electricity connection requests. Among the options put forward is a requirement for data centres to bear the full cost of any network capacity expansions they generate.

Certain operators accept reduced consumption during high-demand intervals, while others position themselves alongside power generation assets. Proposed federal legislation, H.R. 9340, would encourage states to adopt comparable cost-distribution frameworks.

Power companies are working to procure turbines and transformers as worldwide supply constraints continue to affect the energy sector. Restrictions on equipment sourced from China have added pressure to an existing 15 percent shortfall in transformer supply. PPL and Blackstone have formed a joint venture to acquire 5 GW of gas turbines for use in Pennsylvania-based energy projects.

Community pushback has mounted across several states, with local residents raising concerns about noise levels, water consumption, and rising utility costs. Bans on new development have been enacted across upwards of 300 jurisdictions. Lawmakers on both sides of the aisle have raised questions about whether data centres are afforded favourable regulatory treatment.

One gigawatt of generating capacity can supply power on a scale comparable to the energy demands of a medium-sized urban area. In May, the North American Electric Reliability Corporation revised its summer 2026 demand projections downward, attributing the change to grid interconnection holdups and a slower-than-anticipated pace of load growth. Annual peak electricity demand is projected to climb by 11 GW, with data centres accounting for the bulk of that growth.

Demand-response schemes enable certain grid operators to reduce electricity consumption at data centres, cutting projected peak loads in regions such as ERCOT by as much as 3.7 gigawatts. Projections from Bank of America analysts put the country's additional generation capacity requirement at more than 230 GW over the coming five years. Utility providers have scheduled roughly 93 GW of new capacity, while data centres alone are projected to draw an additional 125 GW, pushing annual electricity demand growth to around 4.1 percent.

Construction has begun on data centre developments with on-site power generation totalling over 7.5 GW in capacity. A further 60 GW remains at the pre-construction phase, while AI model sizes continue to expand and inference workloads broaden in scope. Order backlogs at cloud providers have held steady over an extended period.

Forecasts of electricity consumption involve inherent uncertainty, with research pointing both to tendencies toward overestimation in grid connection pipelines and to the potential hazards of building too little capacity. Collaboration between large-scale cloud providers, energy utilities, regulatory bodies, and state governments is widely regarded as a deciding element in the outcome. Options currently being weighed include on-site power generation, adaptable demand management, and faster approval processes for transmission infrastructure.