S&P Global projects that data-center power demand in the United States will roughly triple by 2030, driven largely by the rapid expansion of artificial intelligence workloads. That forecast lands at a moment when grid operators, utilities, and policymakers are already struggling to keep pace with electricity demand that had been flat for more than a decade. Independent data from the International Energy Agency puts global data-center electricity consumption at about 415 TWh in 2024, equal to roughly 1.5 percent of total world electricity use, and projects that figure could reach 945 TWh by 2030 under its base case scenario. The gap between those demand curves and the speed at which new generation and transmission can come online is the central tension shaping electricity markets for the rest of this decade.
Why a tripling of data-center load changes the grid calculus
The scale of the projected increase matters because it is arriving faster than most utility planning cycles can absorb. For years, U.S. electricity demand grew slowly or not at all. Now, data-center expansion alone is expected to account for about 50 percent of electricity demand growth in the United States through 2030, according to the IEA’s Electricity 2026 outlook. That single sector is rewriting load forecasts that utilities use to plan generation, transmission, and distribution investments years in advance.
Average U.S. electricity consumption is projected to rise close to 2 percent per year over the forecast period, a rate that sounds modest until it compounds against a grid that was built for near-zero growth. Regions where hyperscale data centers cluster, including northern Virginia, central Texas, and parts of the Midwest, face the sharpest pressure. New gas plants, solar farms, battery storage, and grid upgrades all require permits, interconnection agreements, and construction timelines that routinely stretch three to five years or longer. In many cases, the lead time for a large transmission line can run even longer than the time it takes to build the data centers that will rely on it.
This is where S&P Global’s tripling projection collides with physical reality. If permitting and transmission bottlenecks slow new generation more than current models assume, the result will not simply be unmet demand. It will show up as higher wholesale electricity prices in constrained regions, reliability warnings during peak periods, and growing tension between data-center operators and existing ratepayers who share the same grid. Neither S&P’s nor the IEA’s central forecasts fully price in the risk that supply-side delays could turn an orderly ramp into a disorderly scramble, especially in markets that already face congestion and aging infrastructure.
IEA data anchors the demand projections
The strongest independent check on S&P Global’s forecast comes from the IEA, which tracks data-center energy use through its own modeling framework. The agency’s base case projects global data-center electricity consumption growing at roughly 15 percent per year from 2024 to 2030, reaching approximately 945 TWh by the end of the decade. That represents more than a doubling from the 2024 baseline of about 415 TWh, and it aligns directionally with S&P’s tripling estimate when the U.S. share of global capacity and the faster domestic growth rate are factored in.
The IEA’s numbers rest on equipment-level analysis of servers, accelerator chips, cooling systems, and networking gear, giving them a bottom-up foundation that pure demand extrapolations lack. AI training runs and inference workloads are the primary drivers of incremental consumption. A single large language model training cluster can draw as much power as a small city, and the number of such clusters is growing rapidly as technology companies race to build out AI infrastructure. As more applications shift from traditional cloud computing to AI-augmented services, the baseline load for inference is likely to rise as well.
For the United States specifically, the IEA’s demand outlook shows overall electricity consumption rising close to 2 percent annually, with data centers responsible for roughly half of that growth. That concentration means the national average understates the local impact. A utility serving a region with multiple new hyperscale campuses could see load growth of 5 to 10 percent or more in a single year, far beyond anything its existing resource plan anticipated. In practice, that forces utilities to revisit integrated resource plans, re-run capacity adequacy studies, and negotiate new power purchase agreements on compressed timelines.
Permitting delays and price risk remain the open questions
Several critical variables sit outside the boundaries of both S&P’s and the IEA’s central projections. The most consequential is the speed of new generation and transmission buildout. Interconnection queues at regional grid operators have ballooned in recent years, with thousands of proposed projects waiting years for approval. If even a fraction of the data-center capacity now under contract or construction comes online before matching generation does, localized supply shortfalls become likely. In those circumstances, system operators may lean more heavily on existing fossil plants, extending their run times and complicating decarbonization targets.
Efficiency gains represent another unresolved factor. Chip designers are improving performance per watt with each generation of AI accelerators, and liquid cooling systems can cut facility-level energy use compared with traditional air cooling. Software optimization and load management tools may also smooth peaks and reduce wasted computation. But history shows that efficiency improvements in computing tend to be offset by rising total workloads, a pattern known as Jevons paradox. Whether this cycle repeats with AI hardware will determine whether actual consumption lands closer to the IEA’s base case or overshoots it as new applications emerge and model sizes continue to grow.
Siting decisions add a further layer of uncertainty. Data-center developers increasingly seek locations with cheap, abundant power and favorable permitting environments. That search is pulling investment toward regions with surplus generation, including parts of the Southeast and areas near nuclear plants. But it is also creating political friction in communities that worry about rising electricity costs, water consumption for cooling, and the visual and environmental footprint of large industrial campuses. Local opposition can translate into new zoning rules, delayed approvals, or conditions that raise project costs and shift development elsewhere.
For utilities and regulators, the open questions around timing, efficiency, and siting converge into a single challenge: how to plan for a surge in demand that is both highly uncertain and highly concentrated. Overbuilding generation and transmission risks saddling ratepayers with stranded assets if AI demand underperforms the most aggressive forecasts. Underbuilding, by contrast, risks reliability events and price spikes that can erode public trust in the energy transition. Navigating between those extremes will require more granular data on data-center load profiles, closer coordination between developers and grid planners, and regulatory frameworks that can adapt faster than in the past.
What is clear from both S&P Global’s projections and the IEA’s modeling is that data centers have moved from a niche load category to a central driver of electricity demand. Over the rest of this decade, the pace at which grids can accommodate that shift-through new generation, expanded transmission, and smarter efficiency measures-will help determine whether the AI boom proceeds on a stable foundation or collides with the physical limits of the power system.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
