The U.S. power grid is being reshaped by a force that most Americans never see: the explosive growth of data centers built to feed artificial intelligence. Federal forecasters now project the strongest four-year surge in electricity demand since 2000, driven almost entirely by large computing facilities. At the same time, major technology companies are investing billions of dollars in private power plants to sidestep the public grid altogether, raising hard questions about reliability, emissions, and who ultimately pays for the infrastructure AI requires.
Federal Forecasts Signal a Demand Spike Not Seen in Decades
The scale of what is coming into focus is difficult to overstate. The U.S. Energy Information Administration has outlined the strongest four-year growth in national electricity demand since 2000 in a recent demand outlook, with that near-term growth extending through 2027. The agency ties the surge explicitly to “large computing facilities, including data centers,” a category that captures the server farms powering generative AI training runs, cloud computing, and enterprise workloads. The projection suggests that what once looked like a temporary spike in consumption is hardening into a structural shift in how and where electricity is used.
The EIA’s Short-Term Energy Outlook breaks the trend down by sector, showing commercial and industrial electricity sales climbing as data center contracts come online across multiple regions. That growth is layered on top of electrification in transportation and buildings, compounding the challenge for utilities that must secure new generation, transmission, and distribution capacity on tight timelines. Instead of the slow, predictable load increases that planners have historically assumed, grid operators are now contending with individual projects that can add hundreds of megawatts of demand in a single stroke.
Regulators Race to Rewrite Grid Rules
Federal agencies are beginning to treat this demand wave as a systemic risk rather than a niche technology story. The Department of Energy, drawing on work from Lawrence Berkeley National Laboratory, released an assessment concluding that electricity load growth has tripled over the past decade and warning that data center demand alone could double or triple by 2028. In that analysis, DOE framed the findings as a national infrastructure challenge, arguing that unmanaged growth in computing load could strain transmission corridors, delay decarbonization, and undermine reliability unless utilities and regulators adjust their planning assumptions. The report’s framing effectively elevates data centers to the same policy tier as electric vehicles and building electrification in long-range grid modeling.
The Federal Energy Regulatory Commission is moving in parallel to update the rules governing how these facilities plug into regional power systems. In an order directed at PJM, the nation’s largest regional transmission organization, FERC instructed the operator to develop new tariff provisions for large computing loads, including data centers that are directly connected to power plants. According to a FERC fact sheet, the commission is especially focused on co-located facilities that draw electricity before it reaches the broader transmission network, because such arrangements can allow big customers to bypass the cost-sharing mechanisms that fund grid upgrades. Regulators worry that if too many tech companies take this route, remaining households and small businesses will be left to shoulder an outsized share of infrastructure costs.
Silicon Valley Builds Its Own Power Supply
Those concerns are not hypothetical. Major technology firms are already pouring money into private generation to keep up with AI workloads, creating what amounts to a parallel electricity system that sits just outside the reach of traditional regulation. Behind-the-meter power plants tied directly to data centers allow companies to avoid multi-year interconnection queues and reduce their exposure to congestion on aging transmission lines. They also give corporate buyers more control over the type of generation they rely on, whether that means gas turbines that can run around the clock or dedicated renewable projects paired with battery storage.
Yet the strategy introduces new vulnerabilities for the broader energy system. Private plants serving data centers are not necessarily integrated into the dispatch tools regional operators use to balance supply and demand, making it harder to call on that capacity during grid emergencies. Because these facilities sit outside many conventional reporting frameworks, their generation and emissions may not be fully reflected in federal tracking systems such as the EPA’s eGRID database, which links plant-level output to carbon intensity across grid regions. That opacity complicates efforts to verify corporate claims about clean energy use and makes it more difficult for policymakers to understand how AI-driven demand is affecting national emissions trajectories.
States Begin Writing Data Centers Into Long-Range Plans
With federal regulators still shaping the contours of national policy, some states are moving ahead with their own planning frameworks that explicitly account for data center growth. In California, the Energy Commission’s latest demand-side forecast for the Integrated Energy Policy Report incorporates data center load as a distinct category alongside electrification trends and behind-the-meter solar. By embedding computing demand directly into the state’s forecast models, planners can adjust transmission investment, resource adequacy requirements, and procurement mandates with a clearer picture of how AI infrastructure will shape future peaks.
California’s approach underscores how far many other jurisdictions still have to go. While national statistics such as the EIA’s natural gas supply data show fuel use rising at gas-fired plants that serve new load, few state-level resource plans break out data centers as a separate driver of demand. In practice, that means large computing projects can appear in forecasts as a diffuse uptick in commercial consumption rather than as concentrated, location-specific loads that may require targeted grid upgrades. Without more granular modeling, utilities risk underestimating the scale of transmission expansions and flexible generation needed to accommodate clusters of AI facilities, especially in regions that are also pursuing aggressive electrification and renewable energy targets.
Who Pays for AI’s Power Hunger?
Underlying all of these developments is a basic question of cost allocation: who ultimately pays for the wires, substations, and backup plants that make AI possible? If data centers connect directly to the grid under traditional tariffs, they contribute to the revenue that funds system-wide investments, but they also accelerate the need for new infrastructure that can push rates higher. If they instead build private plants or negotiate bespoke arrangements with generators, they may avoid some charges altogether, shifting more of the burden onto smaller customers who lack the leverage to secure similar deals. Regulators at FERC and in state commissions are now grappling with how to update interconnection rules, capacity markets, and rate structures so that the beneficiaries of AI-driven growth bear an appropriate share of the costs.
The stakes extend beyond monthly bills. Decisions made over the next few years will influence whether the AI buildout locks in decades of additional fossil fuel capacity or accelerates investment in low-carbon resources. They will determine whether private power islands proliferate or remain the exception, and whether data centers become partners in grid reliability or insulated loads that complicate emergency operations. As federal forecasts, regulatory proceedings, and state planning documents converge on the same conclusion, that AI is now a first-order driver of electricity demand, the policy challenge is shifting from recognizing the problem to designing rules that keep the grid reliable, rates affordable, and emissions on a downward path even as the digital economy’s appetite for power continues to grow.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
