➤ Key Highlights
Utilities across multiple states are overstating AI/data-center electricity demand, using aggressive projections to justify large new generation and transmission spending.
Regulators are pushing back, with Georgia PSC staff recommending cutting projected data-center load by up to 25% due to unrealistic assumptions.
Independent analysis shows structural upward bias in utility forecasts, including assumptions about semiconductor supply growth that global chip producers cannot meet.
Overbuilding based on inflated forecasts would shift financial risk to ratepayers, locking in unnecessary long-lived assets and higher long-term energy costs.
AI-driven load growth is real but volatile, and the regulatory trend is moving toward more conservative modeling and increased scrutiny of claims tied to hyperscale expansion.
U.S. utilities are leaning heavily on aggressive projections of AI-driven electricity demand to justify large-scale grid expansion. Regulators, analysts, and watchdog groups are beginning to challenge those forecasts, warning that many assumptions behind the numbers appear speculative — and could leave ratepayers absorbing the cost of unneeded infrastructure.
The Surge in Data-Center Load Forecasts
Across multiple states, utilities have submitted plans suggesting that data-center development will double or even triple local electricity consumption within just a few years.
• Georgia Power is among the most visible cases, requesting approval for ~9,000 MW of new capacity by 2031, attributing much of that to anticipated data-center demand.
• Nationwide, the narrative is similar: AI compute growth is cited as the primary justification for new generation, transmission, and long-term capital spend.
Regulators are not fully convinced.
🔎 Regulators Push Back on Load Assumptions
During recent proceedings before the Georgia Public Service Commission, staff analysts warned that the utility’s projections may be materially overstated.
Their recommendation: reduce the assumed future data-center load by up to 25%.
The concern is straightforward: utilities may be building long-lived generation assets on the basis of demand that may never materialize.
A parallel report from environmental and policy groups further reinforces this skepticism, noting:
Many projected data centers remain speculative or unconfirmed.
Forecasts rely on a level of semiconductor and server-fabrication output that global suppliers cannot realistically deliver at the pace utilities assume.
The modeling tends to systematically bias demand upward, creating structural incentives to overbuild.
If those campuses fail to materialize, the excess capacity — and its cost — gets pushed onto ratepayers.
The Scale of the Exposure
U.S. data centers consumed ~176 TWh in 2023, roughly 4.4% of national electricity demand. Projections for 2028 range from 6.7% to 12%, depending on the aggressiveness of the modeling.
The spread itself highlights the uncertainty: forecasts are diverging faster than actual construction is occurring.
The risk profile is asymmetric:
Overbuilding locks in higher rates, stranded assets, and long-term fossil-fuel commitments.
Underbuilding creates reliability constraints but can be corrected with incremental investment and short-term capacity tools.
Regulators increasingly prefer the second risk over the first.
The Policy and Investment Implications
Targeted questions are now emerging across the country:
Are utilities using AI/data-center hype to justify capital expansion they already wanted?
A recurring issue flagged by consumer advocates.Should developers bear a greater share of grid-upgrade costs?
Especially in markets where data-center clusters create localized transmission stress.Will excessive load forecasts push utilities toward new gas-fired plants that outlive real demand?
This is one of the core environmental and financial objections raised.Is the current regulatory model even equipped to evaluate AI-linked demand cycles?
Compute growth is highly volatile, tied to hardware availability, capital markets, and hyperscaler strategy — not linear demand fundamentals.
Why It Matters for Investors, Developers, and Operators
The credibility of utility load forecasts affects:
Grid-upgrade timelines
Capital-spending authorization
Wholesale pricing trajectories
Siting feasibility for energy-intensive facilities
Long-term policy direction
For real-estate operators (including industrial, hyperscale, and mixed-use), these dynamics determine both the cost structure and the development timetable of large projects.
If regulators force utilities to recalibrate their assumptions downward, the next decade of grid investment could shift substantially — with consequences for every asset class that depends on reliable, affordable power.
Stop Reading Headlines
Start Understanding the Market
➤ TAKEAWAY
The U.S. power sector is entering a phase where AI-driven demand forecasts have outsized influence on capital allocation, yet the underlying assumptions remain fragile. Utilities have strong structural incentives to over-predict, while regulators are increasingly skeptical of those claims.
The result is an emerging tension between infrastructure ambition and demand realism, with ratepayers exposed to the downside if projections fail to materialize.
Anyone developing, financing, or operating energy-dependent real estate should expect more conservative regulatory oversight, slower approval cycles, and heightened scrutiny of “AI load” claims over the coming years.
