How Are Data Centers and AI Infrastructure Reshaping Utility Load Growth and Investment Opportunity?

Data center electricity demand represents the most significant change in US utility load growth since the electrification of industry and housing in the 20th century — potentially reversing two decades of flat-to-declining electricity demand in a sector that had been structurally challenged by energy efficiency improvements and industrial offshoring. The convergence of artificial intelligence training and inference workloads (GPU clusters consuming 10–100 megawatts each), hyperscaler cloud infrastructure expansion (Microsoft, Amazon, Google, Meta each planning hundreds of billions in data center investment), and edge computing proliferation creates electricity demand growth in specific utility service territories that is transforming the investment thesis for affected utilities from "slow-growth income" to "genuine growth story."

Quick definition: Data center electricity demand terms: (1) Power Usage Effectiveness (PUE) — data center efficiency metric; ratio of total facility power to IT equipment power; best-in-class 1.1–1.2; older facilities 1.4–1.7; (2) Hyperscaler — mega-scale cloud and AI infrastructure operators (Microsoft Azure, Amazon AWS, Google Cloud, Meta); each consumes gigawatts of electricity annually; (3) Colocation — third-party data center facilities housing servers for multiple tenants; major operators include Equinix, Digital Realty; (4) Campus demand — hyperscaler single-campus deployments of 500 MW to 2+ GW in a single location; (5) Load letter of intent — utility customer formal commitment to connect; disclosure of signed letters of intent provides visibility into forward load growth.

Key takeaways

• Dominion Energy serves Northern Virginia, which hosts approximately 70% of all US internet traffic and houses the world's largest concentration of data center capacity — Northern Virginia's data center load exceeded 3.5 gigawatts in 2024 and is projected to reach 8–12 gigawatts by 2030, requiring Dominion to build transmission and generation at a pace comparable to constructing an entirely new utility from the region's existing load
• The utility-hyperscaler dynamic creates a new category of large commercial customer relationship — Microsoft, Amazon, and Google negotiate directly with utilities for power purchase agreements, transmission connection commitments, and carbon-free energy guarantees; these hyperscaler customers are willing to pay premium rates for guaranteed capacity and zero-carbon power attributes, changing the rate design dynamics in data center hub markets
• Transmission investment driven by data center load growth is FERC-regulated, earning 10–11% allowed ROE with potential incentive adders — utilities building transmission to serve data center hub markets earn federally-regulated returns on investment, providing attractive regulated earnings from infrastructure buildout that serves both data center customers and broader grid modernization
• The supply-constrained nature of data center electricity service — grid connection queues stretching 4–8 years in Virginia and other hub markets, transformer lead times of 24–48 months, transmission permitting of 5–10 years — creates a durable competitive moat for utilities already positioned in high-density data center markets; entry by alternative power providers is nearly impossible at the timescale hyperscalers require
• Duke Energy Carolinas has identified over 6 GW of data center load growth interest in North Carolina's Research Triangle and Charlotte corridor — from Apple, Google, Meta, and numerous other technology companies expanding their southeastern US data center footprints; this load growth would approximately double Duke Carolinas' peak demand within 10 years, transforming a mature regulated utility into a high-growth infrastructure business

The AI electricity demand thesis

Generative AI computational requirements: Large language model training — creating GPT-scale and larger AI models — requires enormous computational resources. A single large training run can consume millions of GPU-hours of compute, with each GPU cluster consuming 1–10 megawatts continuously for months. As AI model complexity increases and inference (running trained models at scale for user queries) proliferates, electricity demand grows beyond the initial training phase. Goldman Sachs estimated in 2024 that data centers could consume 8% of total US electricity by 2030, up from approximately 3% in 2023 — representing approximately 250 terawatt-hours of incremental annual demand.

Hyperscaler capital commitment scale: Microsoft announced $80 billion in data center investment for fiscal year 2025; Amazon has guided $75–80 billion in annual capex with data center investment as the primary component; Google announced $75 billion in 2025 capex plans including significant data center expansion; Meta announced $60–65 billion in capex with AI infrastructure as the primary use. These capital commitments translate directly into electricity demand — every dollar of data center construction becomes an electricity consumer with multi-decade operating life.

Geographic concentration in utility service territories: Data center geography is driven by land availability (large tracts of flat land for campus construction), water availability (for cooling), fiber connectivity (for low-latency internet interconnection), and — most importantly from a utility perspective — available electricity capacity with existing transmission infrastructure. Northern Virginia emerged as the dominant US data center market because of these factors plus proximity to federal government cloud customers. The mid-Atlantic PJM grid (Dominion's territory) hosts this concentration; other data center corridors are developing in Oregon (Pacific Northwest utilities), Texas (ERCOT utilities, particularly AEP Texas), and North Carolina/Georgia (Duke Energy, Georgia Power/Southern Company).

How it flows

Dominion Energy Northern Virginia analysis

Scale of the demand signal: Dominion Energy's 2023 IRP (Integrated Resource Plan) projected Northern Virginia data center load growing from approximately 3.5 GW in 2023 to 6–9 GW by 2035 — requiring construction of approximately 3–6 GW of new generation capacity plus extensive transmission infrastructure to serve this demand. For context, this load growth projection is equivalent to adding the entire electricity demand of a mid-sized state within Dominion's existing service territory over 10–12 years.

Transmission investment requirements: Serving data center campuses (500 MW to 2 GW each) requires new 500 kV or 765 kV transmission lines — the highest-voltage infrastructure in the US grid. These transmission projects require FERC and state environmental permitting, typically 5–10 years from concept to energization. Dominion has identified over $20 billion in transmission investment requirements for Northern Virginia load growth alone — FERC-regulated at 10–11% allowed ROE with potential transmission incentive adders.

Carbon-free power requirement: Microsoft, Amazon, Google, and Meta have each committed to matching 100% of their electricity consumption with carbon-free energy — driving hyperscaler demand specifically for nuclear, wind, solar, and hydroelectric power rather than fossil fuel generation. This carbon-free requirement creates pressure for Dominion and other utilities to accelerate renewable energy development alongside conventional generation. Dominion's offshore wind development (Coastal Virginia Offshore Wind — 2.6 GW approved capacity) serves both grid reliability needs and hyperscaler carbon-free requirements simultaneously.

Duke Energy Carolinas opportunity

Research Triangle and Charlotte data center development: Apple's data centers in Maiden, NC; Google's data centers in Lenoir, NC and Berkeley County, SC; Meta's facility in Catawba County, NC — and multiple additional hyperscaler facilities in development across the Charlotte metro area and Raleigh/Durham Research Triangle — represent a data center corridor in Duke Carolinas' territory that is just beginning to scale relative to Northern Virginia's more mature market.

Load growth momentum indicators: Duke Energy tracks large customer load letters of intent (LOIs) as a leading indicator of load growth. Management's 2024 disclosure of approximately 6 GW of data center LOIs in the Carolina service territory — from companies at various stages of site selection and development — provided unprecedented visibility into potential load growth. Converting even 50% of this LOI pipeline into actual connections by 2030 would double Duke Carolinas' peak demand, requiring massive capital investment with NCUC regulatory approval.

North Carolina regulatory support: North Carolina's Public Utilities Commission has historically been constructive — approving Duke Energy's capital investment programs and supporting economic development through reliable electricity infrastructure. The data center growth story aligns political objectives (job creation, tax revenue, technology industry attraction) with utility capital investment — creating political tailwinds for regulatory approval of Duke's data-center-driven capital program.

Transmission grid constraint as competitive moat

Queue congestion reality: PJM's interconnection queue (the line of projects waiting to connect to the grid) has grown to over 300 GW of projects waiting for connection studies — with data center and generation projects queuing simultaneously. Average interconnection study timelines have stretched from approximately 2 years to 4–8 years in some regions. This queue congestion means that hyperscalers seeking new electricity connections face years-long delays in markets without existing transmission capacity — creating durable demand for utilities that can accelerate connections through existing infrastructure.

Equipment supply chain constraints: Transformers (particularly large power transformers for data center connections) have lead times of 24–48 months in the current market — up from 12 months before COVID and the clean energy buildout concurrent demand surge. High-voltage switchgear, protective relays, and substation equipment face similar elongated lead times. Utilities with existing inventory and supplier relationships can serve data center customers faster than new entrants — another structural moat element.

Merchant versus regulated opportunity: Merchant power producers (Constellation Energy, Talen Energy) have pursued direct nuclear PPAs with hyperscalers — Amazon's acquisition of Talen Energy's Susquehanna nuclear plant campus (Cumulus Data) and Constellation's Microsoft TMI restart agreement demonstrate that hyperscalers are willing to invest directly in power infrastructure to secure long-term carbon-free supply. This merchant approach supplements rather than competes with regulated utilities — hyperscalers need both on-site generation and grid transmission to serve distributed global infrastructure.

Utility data center exposure identification

IRP load growth disclosures: State-filed Integrated Resource Plans are the primary source for utility-disclosed data center load growth projections. IRP filings (typically every 2–3 years) include demand forecasts by customer class, load growth assumptions, and large customer pipeline disclosures. Reviewing the most recent IRP filing for each utility and comparing current load growth projections to prior IRP projections reveals which utilities have dramatically revised upward their demand forecasts based on data center pipeline.

Letter of intent and executed contract disclosures: Utilities disclose signed large-customer LOIs and executed service agreements in earnings calls and investor presentations. Management teams that explicitly quantify data center load pipeline (in gigawatts and by timeline) provide superior visibility for modeling load-growth-driven capital program expansion.

Transmission planning documents: PJM, MISO, WECC, and SERC publish regional transmission planning documents that identify major load additions driving transmission investment needs. These documents (available at each regional transmission organization's website) quantify the transmission investment required by region and identify which utility service territories are experiencing the most rapid load growth.

Common mistakes

Assuming all utilities benefit equally from data center growth. The data center electricity demand story is highly geographically concentrated — Northern Virginia, North Carolina Research Triangle, Oregon/Washington Pacific Northwest, Texas (ERCOT), Georgia/South Carolina are the primary beneficiary markets. Utilities without data center hub exposure (rural Midwest utilities, New England utilities, Gulf Coast utilities) will not experience meaningful load growth from this trend. Buying the entire utility sector as a "data center play" dilutes exposure to the handful of genuinely impacted utilities.

Extrapolating current hyperscaler capital commitments linearly. Hyperscaler data center investment can slow if AI revenue growth disappoints expectations, if efficiency improvements reduce per-inference computing requirements, or if capital allocation priorities shift. Utility investors who have priced in maximum data center load growth scenarios face downside risk if actual load materializes more slowly than pipeline LOIs suggest. Load letters of intent are non-binding; LOI-to-actual-connection conversion rates of 50–70% are more realistic than 100%.

FAQ

How do data center customers' zero-carbon requirements affect utility capital programs and rate design?

Hyperscaler requirements for carbon-free power create a specific demand signal for renewable energy and nuclear generation that changes utility IRP planning and rate design. Traditional utility rate design charges customers for generation, transmission, and distribution based on their load share — without distinguishing carbon content. Hyperscaler preferences are driving new rate structures: green tariffs (enabling customers to pay a premium for verified renewable energy allocation), direct renewable PPAs with utility as facilitator, and behind-the-meter renewable generation with grid backup. Virginia, North Carolina, and other states have enacted specific legislation enabling utility green tariffs for large commercial customers — Virginia's SCC and North Carolina's NCUC have approved programs for Dominion and Duke respectively. These green tariff programs generate incremental renewable energy investment that earns regulated returns — creating an alignment between hyperscaler carbon goals, utility capital investment, and regulatory approval. The EIA Electric Power Monthly tracks electricity sales by state and customer class at eia.gov; PJM's annual load forecast incorporates data center demand at pjm.com.

Related concepts

• Electric Utilities
• Utilities Economic Cycle
• Utilities Historical Performance
• Utilities Regulation
• Utilities Portfolio Sizing

Summary

Data center electricity demand is transforming select utilities from slow-growth income investments to high-capital-program growth businesses — most acutely for Dominion Energy (Northern Virginia, 3.5 to 8–12 GW projected by 2030), Duke Energy Carolinas (6 GW of data center LOIs in Research Triangle/Charlotte), Georgia Power (Southern Company in the Southeast corridor), and AEP Texas (ERCOT data center growth). The AI infrastructure buildout (Microsoft $80B, Amazon $75–80B, Google $75B, Meta $60–65B in annual capex) creates decade-scale electricity demand growth in specific utility service territories. Grid connection queue congestion (4–8 years in constrained markets), transformer lead times (24–48 months), and transmission permitting complexity (5–10 years) create a durable competitive moat for established utilities in data center hub markets — preventing rapid alternative entry. Carbon-free power requirements drive green tariff and renewable energy programs that generate additional regulated capital investment. The thesis is geographically concentrated — utilities outside data center hub markets do not benefit — requiring selective individual company selection rather than broad utility sector exposure.

Next

→ Utilities Portfolio Sizing: Rate Cycle, Defensive Allocation, and Positioning Framework