Powering the AI Boom: Can the U.S. Expand Data Centers Without Raising Electricity Costs?

AI’s Power Demand Is Testing the U.S. Grid

Artificial intelligence may be a digital industry, but its rapid expansion is creating an increasingly physical problem: electricity. The data centers required to develop and operate AI systems consume enormous amounts of power, and the speed at which technology companies want to build them is beginning to test the infrastructure needed to supply it. Lawrence Berkeley National Laboratory estimates that data centers could account for 11.8% of total U.S. electricity consumption by 2030, with projections ranging from 9.5% to 15.3% depending partly on the pace and intensity of AI computing growth (Smith et al., 2026). For an electricity system that experienced relatively little demand growth for much of the previous two decades, the shift is significant.

Federal regulators are now trying to determine how the grid should adapt. On June 18, the Federal Energy Regulatory Commission (FERC) issued six orders directing the country’s federally regulated regional grid operators to justify or reform the rules governing how data centers and other large electricity users connect to the transmission system (FERC, 2026). The policy is attempting to accomplish two potentially competing objectives: give large users faster access to electricity while preventing the infrastructure costs associated with serving them from being shifted onto existing customers. As investment in AI infrastructure accelerates, what appears to be a technical debate over electricity regulation is becoming a broader question about who pays for America’s AI boom.

Why FERC Is Changing the Rules

The existing electricity system was not designed for the scale or speed of today’s data-center expansion. Unlike gradual increases in household or commercial electricity consumption, individual data centers can introduce enormous amounts of demand in a single location. At the same time, connecting new generation and transmission infrastructure can take years, creating a mismatch between the pace of technology investment and the pace at which the electricity system can expand.

FERC’s June action is intended to address that mismatch. Rather than immediately imposing a single national framework, the Commission issued tailored orders to PJM, MISO, SPP, CAISO, ISO New England, and NYISO. Each operator was given 60 days to either demonstrate that its existing tariffs remain reasonable or propose changes addressing five areas identified by FERC (FERC, 2026). These include faster transmission-service studies, greater transparency around infrastructure costs, rules for data centers located alongside their own generation, new transmission services for large customers capable of reducing consumption when necessary, and processes for studying generation built near large loads.

Cost allocation is at the center of the policy. Connecting a large data center may require new transmission lines, substations, generation, or other equipment. If those investments are incorporated broadly into electricity rates, households and smaller businesses could end up paying for infrastructure built partly to serve some of the world’s largest technology companies. FERC is therefore seeking to prevent inappropriate cost shifting while still providing the regulatory certainty needed for large-scale investment (FERC, 2026). The policy is not intended to slow the AI buildout. Instead, it attempts to change the incentives around how new data centers connect to and potentially support the grid.

AI’s Electricity Problem Is Becoming an Economic Problem

The scale of projected demand explains why regulators are acting now. Lawrence Berkeley’s reference case projects U.S. data centers consuming roughly 649 terawatt-hours of electricity in 2030, but its broader estimates range from 521 to 843 TWh depending on factors including AI-chip deployment and server utilization (Smith et al., 2026). That uncertainty creates its own challenge. Grid planners must prepare for rapidly rising demand without knowing exactly how many proposed facilities will ultimately be built or how intensively they will operate.

The shift is already visible in national electricity consumption. U.S. electricity demand grew by an average of just 0.1% annually between 2005 and 2019, but that rate increased to 1.7% between 2020 and 2025, with data centers helping drive the acceleration (EIA, 2026). The EIA expects electricity load to rise another 1.9% in 2026 and 2.5% in 2027, with particularly strong growth in regions including Texas and PJM (EIA, 2026).

The economic consequences depend heavily on whether electricity supply can expand alongside that demand. New power plants and transmission infrastructure often require long planning, construction, and interconnection timelines. If demand instead grows faster than expected in the near term, the EIA finds that grid operators would have to rely more heavily on existing natural gas and coal generation. In its high-demand scenario, the consequences vary substantially by region: modeled 2027 wholesale electricity prices in PJM rise roughly 4% above the agency’s baseline forecast, while prices in the more isolated ERCOT market rise 79% (EIA, 2026). These are scenarios rather than predictions, but they demonstrate how quickly the economics can change when demand grows faster than available supply.

This is what turns the data-center boom from a technology story into a broader economic issue. Higher wholesale electricity costs can eventually affect households and businesses, while grid constraints can delay billions of dollars in private investment. FERC therefore faces a difficult balancing act: insufficient infrastructure risks slowing AI development, but rapidly expanding the grid without carefully allocating those costs risks making consumers finance part of that expansion.

The Bubble Question

That balancing act is getting harder. FERC has not imposed a single nationwide framework governing what data centers must pay to connect to the grid, but its June orders make clear that the status quo is under review. At the same time, political pressure is growing as communities confront the local consequences of an investment boom that, until recently, was largely a Wall Street story.

That leaves FERC in an awkward position, because the same construction boom generating public backlash has also become a major source of U.S. capital investment. Technology companies are committing enormous sums to data centers, computing equipment, and other AI infrastructure, making access to electricity increasingly important not only for the technology sector but for the broader investment outlook. Slowing the buildout could therefore constrain one of the country’s fastest-growing areas of capital spending.

This raises an uncomfortable question: how much of the AI boom is justified by what AI can already do, and how much by what investors believe it will eventually do? Today's valuations are built heavily around expectations of future earnings. Goldman Sachs estimates that companies associated with the AI theme have added roughly $27 trillion in market value since late 2022, although that increase cannot be attributed entirely to AI and its justification still depends heavily on expectations of future earnings growth. The infrastructure spending is part of that expectation. More models, more users, and more valuable AI products require more computing capacity, which requires more chips, more electricity, and more data centers.

That makes the data-center buildout both the foundation of the AI investment thesis and one of its greatest vulnerabilities. If investors begin to doubt that the infrastructure can be built quickly enough to support the capabilities they are pricing in, the valuations resting on it become considerably harder to defend. The public is increasingly demanding that someone make data centers pay their own costs; investors are increasingly assuming that those data centers will be built. FERC's June orders do not resolve that conflict, but they make electricity availability and infrastructure cost allocation increasingly relevant variables in the AI investment thesis. If power constraints materially increase the cost or delay the construction of new capacity, the returns assumed from current levels of AI capital expenditure become harder to achieve.

Can America Win the AI Race Without Making Consumers Pay for It?

The debate ultimately reflects a larger shift in U.S. technology policy. Competing in artificial intelligence is usually framed around semiconductors, computing power, investment, and the capabilities of individual models. Yet the physical infrastructure supporting those systems is becoming just as important. A company can have the capital and chips needed for a new data center and still face years of delays if sufficient electricity is unavailable.

FERC's approach attempts to create an alternative to choosing between rapid expansion and consumer protection. Large users that impose substantial infrastructure costs could be required to bear those costs more directly, while projects capable of locating near generation, supplying some of their own power, or reducing consumption during periods of grid stress could receive more flexible access (FERC, 2026). If those incentives work, they could influence not only how quickly data centers connect, but where companies build them and how they secure their electricity.

There is also good reason not to assume that every new data center will automatically raise household electricity bills. The impact depends on regional generating capacity, transmission constraints, utility regulation, and how infrastructure costs are distributed. The sharp difference between the EIA's modeled effects in ERCOT and PJM demonstrates that identical demand growth can produce very different outcomes depending on the structure of the regional grid (EIA, 2026). The policy challenge is therefore less about stopping electricity-demand growth than about ensuring that investment in supply and infrastructure keeps pace with it.

The U.S. can expand data-center capacity without necessarily imposing substantially higher electricity costs on existing consumers, but only if generation and transmission investment keeps pace with demand and the costs created by large new users are allocated appropriately. FERC's June orders are therefore about more than technical changes to electricity tariffs: they reflect the growing recognition that the AI race is becoming an energy and infrastructure race as well. Data centers could consume more than one-tenth of U.S. electricity by the end of the decade, making electricity availability, grid investment, and cost allocation increasingly important determinants of how quickly AI infrastructure can expand. America's ability to compete in artificial intelligence may increasingly depend not only on the technology inside its data centers, but on whether the electricity system outside them can keep up.