For two decades after the commercialization of the internet, U.S. electricity demand was essentially flat. The grid grew at less than 1% annually as efficiency gains offset the rise of the digital economy. Power utilities across the country planned for minimal growth and invested accordingly. Then came generative AI.

The five largest technology companies — Amazon, Alphabet, Microsoft, Meta, and Oracle — collectively plan to spend between $660 and $690 billion on capital expenditure in 2026, a 36% increase from their 2025 levels, which were themselves up 73% from 2024. Approximately 75% of that spending — roughly $450 to $500 billion — is directed at AI infrastructure: GPU clusters, data centers, networking, and power systems. Amazon's capex alone in 2026 is larger than the entire U.S. publicly traded energy sector's combined annual spending on drilling wells, extracting oil and gas, and running refineries.

That is an extraordinary statistic. And it points directly to the problem: an industry scaling at this velocity needs power at a scale the existing grid was never designed to provide.

The Hyperscaler 2026 Commitments: Amazon: ~$200 billion in capex for 2026. Alphabet/Google: $175-185 billion. Microsoft: ~$120 billion or more. Meta: $115-135 billion. Oracle: ~$50 billion. Combined: $660-690 billion. This compares to $256 billion across the same five companies in 2024 — a nearly three-fold increase in two years. Data center capital expenditure across the entire industry is projected to cross $1 trillion in 2026, a milestone analysts had previously estimated would not arrive until 2029.

To understand the magnitude of the power problem, start with a single query. A standard Google search consumes approximately 0.3 watt-hours of electricity. A ChatGPT query consumes approximately 2.9 watt-hours — nearly ten times as much. Training the models that power those queries requires orders of magnitude more. A large-scale AI model training run can consume electricity equivalent to tens of thousands of homes for weeks at a time.

The hardware compounds this. Traditional server racks in data centers typically operate at 7 to 10 kilowatts per rack. AI computing racks — housing Nvidia GPUs and their equivalents — can operate at 30 to 100+ kilowatts per rack. That is a 10-fold increase in power density per unit of physical space. It means AI data centers require not just more buildings but fundamentally different power infrastructure, cooling systems, and grid connections than their conventional counterparts.

The International Energy Agency projects that global data center electricity consumption will reach 1,100 terawatt-hours in 2026 — equivalent to Japan's entire national electricity consumption — and that this represents an 18% upward revision from forecasts made just months earlier. In the United States, the Lawrence Berkeley National Laboratory projects that data center demand will grow from 176 terawatt-hours in 2023 to between 325 and 580 terawatt-hours by 2028 — roughly tripling to quadrupling in five years. The IEA expects global data center demand to more than double to 945 terawatt-hours by 2030, with the U.S. accounting for approximately half of that growth.

Virginia: Ground Zero for the Grid Strain. Northern Virginia is home to the highest concentration of data centers in the world, sometimes called "Data Center Alley." In 2023, data centers already consumed approximately 26% of Virginia's total electricity supply. The state's largest utility, Dominion Energy, proposed its first base-rate increase since 1992 in early 2025 — adding roughly $8.51 per month to a typical household bill beginning in 2026. PJM Interconnection, the regional grid operator, projects it will be 6 gigawatts short of its reliability requirements by 2027, with data center growth identified as the primary driver. Capacity market prices in PJM spiked nearly ten-fold in a single year as a direct result.

The electricity grid in the United States was largely designed and built in the mid-twentieth century. It was built for a world of factories, homes, and offices with relatively predictable and gradual load growth. It was not built for a world in which a single data center campus can demand more power than a medium-sized city, and demand it immediately.

To connect a new large power consumer to the grid, a utility must conduct an interconnection study — assessing whether the existing infrastructure can handle the new load and what upgrades are required. In 2025, these queues became severely backlogged. Texas passed Senate Bill 6 in 2025 specifically to create a new framework for data centers exceeding 75 megawatts, requiring extensive upfront studies and financial commitments before interconnection. In Northern Virginia, the wait for interconnection approval has extended to multiple years, delaying data center projects and forcing some hyperscalers to pursue alternative power strategies.

Building a large-scale data center can be done in 18 to 24 months. Building the power generation and transmission infrastructure to support it takes 5 to 15 years under current regulatory and permitting frameworks. New high-voltage transmission lines require environmental review, local permits, and often litigation. New power plants of any significant scale face similar timelines. This fundamental mismatch between the speed of AI infrastructure deployment and the speed of energy infrastructure development is the core of the grid crisis.

Consumer internet services can tolerate occasional outages. AI training runs cannot. Interrupting a multi-week GPU cluster training job is catastrophically expensive — you lose the entire computation. Data centers therefore require 99.999% uptime, or approximately five minutes of downtime per year. The grid, even in the United States, does not reliably achieve this standard. This is why hyperscalers have been investing aggressively in on-site power generation and storage — not to reduce their grid connection costs, but to ensure the reliability standard that AI operations demand.

The power problem has three near-term solutions and one long-term answer that every major technology company is pursuing simultaneously. None of them individually solves the problem. All of them together define the investment landscape.

Natural gas is currently the dominant answer to the power problem because it can be deployed quickly, operates reliably around the clock, and is available in large quantities. Amazon, Microsoft, and Alphabet have all signed contracts with natural gas suppliers and are developing on-site gas generation capacity at their largest facilities. Gas turbines can be installed in 18 to 24 months — fast enough to keep pace with data center construction in a way that other sources cannot yet match.

This creates a tension with the sustainability commitments all three companies have publicly made. Each has pledged carbon-neutral or carbon-free operations by various dates. Running gas turbines to power AI training runs moves in the opposite direction. The hyperscalers characterize natural gas as a temporary bridge while cleaner alternatives scale up. Whether that framing holds over a decade of continued AI growth is an open question.

Nuclear power has emerged as the preferred long-term solution for one simple reason: it generates large quantities of continuous, carbon-free power with no dependence on weather or time of day. Wind and solar are cheaper on a per-unit basis but are intermittent — the wind does not always blow and the sun does not always shine. A data center running a training job at 3 AM on a calm, cloudy winter night needs reliable power regardless of weather conditions. Nuclear provides it.

The deals being signed are significant. Microsoft entered into a 20-year agreement with Constellation Energy to restart the Three Mile Island nuclear plant in Pennsylvania, targeting 2028 restart at a cost of approximately $1.6 billion. The plant had been shut down since 2019. Google signed the first U.S. corporate Small Modular Reactor fleet deal with Kairos Power for approximately 500 megawatts by 2030 and beyond. Amazon has committed over $20 billion to convert the Susquehanna nuclear plant into an AI campus. Meta issued a request for proposals for 1 to 4 gigawatts of new nuclear capacity and signed a massive power purchase agreement with Vistra Energy for its Comanche Peak nuclear facility.

Goldman Sachs identified Vistra (NYSE: VST) as its top pick in the AI power trade, with a price target of $205. The firm raised its 2027 EBITDA estimate for Vistra by 5% following the Meta deal announcement, noting that the company is securing large, long-term power purchase agreements faster than expected.

Rather than waiting years for grid upgrades, an increasing number of hyperscalers are pursuing on-site power generation — effectively building small power plants adjacent to or integrated with their data center campuses. This includes natural gas reciprocating generators, solar installations, battery storage systems, and in some cases fuel cells. The logic is straightforward: if the grid cannot deliver the reliability and capacity needed, build your own grid segment.

Google acquired Intersect Power, a renewable energy company, for $4.75 billion, specifically to accelerate its ability to develop on-site power generation for its data centers. This acquisition represents a tech giant vertically integrating into energy production in a way that has no historical precedent.

The electricity grid operates as a shared cost infrastructure. When utilities build new transmission lines, upgrade substations, or add generation capacity to serve data centers, the cost is not borne exclusively by data centers — it is spread across all ratepayers in the service territory. This is the traditional utility cost-allocation model that has existed for over a century, and it is generating growing political controversy.

In Ohio, one household — Ken and Carol Apacki — saw their monthly electricity bill rise from $12 to $19 in a single year, a 60% increase, as their region became home to over 130 data centers. Their experience is not isolated. The PJM Interconnection's capacity market clearing prices for the 2026-2027 delivery year were $329.17 per megawatt — more than ten times the price of $28.92 per megawatt in the 2024-2025 delivery year. This tenfold spike in wholesale capacity prices filters through to retail electricity bills. Some service areas within PJM have seen retail rate increases exceeding 15%.

Nationally, retail electricity prices have risen 42% since 2019, significantly outpacing overall CPI inflation of approximately 29% over the same period. Goldman Sachs projects that data center-driven demand will add 0.1 percentage points to core inflation in both 2026 and 2027. A Carnegie Mellon University study estimates that data centers and cryptocurrency mining together could lead to an 8% increase in the average U.S. electricity bill by 2030, with the highest-demand regions — central and northern Virginia — potentially seeing increases above 25%.

The Political Dimension: "The fundamental question is whether middle-class families should subsidize the electricity needs of companies worth trillions of dollars." — Sanya Carley, Professor of Energy Policy, University of Pennsylvania. This question is becoming a live political issue on both sides of the aisle. Utilities have requested $31 billion in rate hikes during 2025 alone. With more elections in 2026 than 2025, energy affordability driven by data center demand is emerging as a genuine campaign issue.

The AI power problem is not just a grid challenge — it is one of the most significant investment themes of the current decade. The capital flows are massive, the timelines are long, and the infrastructure being built will serve as the physical foundation of the AI economy for the next 20 to 30 years.

It is worth noting the distinction between companies with genuine, current-revenue AI power businesses and those trading primarily on the promise of future participation. Small Modular Reactor developers like Oklo (NYSE: OKLO) trade at valuations implying hundreds of millions in earnings that do not currently exist. New-build nuclear costs approximately $200 to $250 per megawatt-hour, and while hyperscalers are contracting at rates that make this viable, the path from early-stage development to commercial revenue is long and uncertain. The better-positioned nuclear infrastructure play, as noted by analysts at Altimetry, may be BWX Technologies (BWXT), which manufactures nuclear reactor components for the U.S. Navy today and carries proven technology, current revenue, and less speculative premium.

The AI power problem is not a theoretical future risk. It is a current operational constraint that is shaping where data centers get built, how they get powered, and how much American households pay for electricity. The grid, built for a different era, cannot scale at the speed the AI industry demands. That gap is being filled by natural gas in the short term and by an unprecedented wave of private investment in nuclear power for the long term.

For investors, the infrastructure buildout behind AI is arguably the more durable investment story than the AI model companies themselves. The companies building the physical layer — the power plants, the transmission lines, the cooling systems, the electrical equipment — are providing a product with genuine scarcity and long-term contracted demand. Unlike the AI model layer, where competition is fierce and the economics of model providers are uncertain, the power layer has clear customers, long-term contracts, and an infrastructure gap that cannot be closed quickly.

The hyperscalers are spending more than $600 billion in 2026 to build the AI era's physical infrastructure. Understanding where that money flows — and who collects it — is as important as understanding which AI company has the best model.

^{_{All sources accessed April 2026. For informational and educational purposes only. Not financial advice.}}

^{_{[1] International Energy Agency (IEA). "Energy and AI." iea.org, April 2025.}}

^{_{[2] Pew Research Center. "What We Know About Energy Use at US Data Centers Amid the AI Boom." pewresearch.org, October 24, 2025. (Citing Lawrence Berkeley National Laboratory and IEA data.)}}

^{_{[3] Belfer Center for Science and International Affairs (Harvard Kennedy School). "AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment." belfercenter.org, February 10, 2026.}}

^{_{[4] NZero / U.S. Energy Information Administration. "U.S. Power Demand Hits New Highs Driven by Data Centers, AI, and Grid Constraints." nzero.com, December 15, 2025. (Citing EIA December 2025 Short-Term Energy Outlook.)}}

^{_{[5] Common Dreams / CNBC. "US Electric Grid Heading Toward 'Crisis' Thanks to AI Data Centers." commondreams.org, January 3, 2026. (Citing PJM Interconnection and Joe Bowring / Monitoring Analytics.)}}

^{_{[6] Tech Insider. "AI Data Centers Now Use More Power Than 30 Countries — The 2026 Crisis." tech-insider.org, updated April 2, 2026. (Citing IEA, Goldman Sachs, PJM, Carnegie Mellon, Brookings data.)}}

^{_{[7] arXiv / Academic Research. "Electricity Demand and Grid Impacts of AI Data Centers: Challenges and Prospects." arxiv.org, September 10, 2025.}}

^{_{[8] NPR. "AI Data Centers Use a Lot of Electricity. How It Could Affect Your Power Bill." npr.org, January 2, 2026.}}

^{_{[9] Morgan Stanley. "Powering AI: Markets Race to Invest in AI Energy Solutions." morganstanley.com, February 27, 2026.}}

^{_{[10] Futurum Group. "AI Capex 2026: The $690B Infrastructure Sprint." futurumgroup.com, February 12, 2026.}}

^{_{[11] Introl Blog. "Hyperscaler CapEx Hits $600B in 2026." introl.com, January 7, 2026. (Citing CreditSights, Morgan Stanley, JPMorgan data.)}}

^{_{[12] IEEE ComSoc Technology Blog. "Hyperscaler Capex >$600bn in 2026, a 36% Increase Over 2025." techblog.comsoc.org, December 22, 2025.}}

^{_{[13] Network World / Dell'Oro Group. "Hyperscaler Backlogs Show Growing Demand for AI Infrastructure." networkworld.com, April 2026.}}

^{_{[14] Morningstar. "AI Arms Race: How Tech's Capital Surge Will Reshape the Investment Landscape in 2026." morningstar.com, December 12, 2025.}}

^{_{[15] Introl Blog. "Nuclear Power for AI: Inside the Data Center Energy Deals." introl.com, January 8, 2026. (Covering Microsoft Three Mile Island, Google Kairos Power, Amazon Susquehanna, Meta nuclear RFP.)}}

^{_{[16] Nasdaq / Data Center Knowledge. "3 Nuclear Power Stocks Set to Flourish in 2026 on AI Data Center Boom." nasdaq.com, December 19, 2025.}}

^{_{[17] Nasdaq / Investing.com. "2 Under-the-Radar Energy Stocks to Watch for AI Demand in 2026." nasdaq.com, December 29, 2025.}}

^{_{[18] Investing.com / Altimetry. "3 Energy Stocks to Buy as AI Power Demand Surges — and 2 to Avoid." investing.com, April 2026.}}

^{_{[19] Investing.com. "Big Tech Will Spend $600B on AI in 2026: 5 Stocks Cashing the Checks." investing.com, February 6, 2026. (Citing Goldman Sachs Vistra analysis.)}}

^{_{[20] Constellation Energy / SEC. "Form 8-K: AI Data Center Power." sec.gov, 2025. (Containing hyperscaler capex quotes from Amazon, Google, and Meta earnings calls.)}}

DISCLAIMER

This article is published by Brezco Analytics for informational and educational purposes only. Nothing contained herein constitutes financial, investment, legal, or tax advice. All information is believed to be from reliable sources but is not guaranteed. Past performance is not indicative of future results. The author may or may not hold positions in any securities discussed. Always conduct your own due diligence and consult a licensed financial advisor before making investment decisions. This is not a solicitation to buy or sell any security.

© 2026 Brezco Analytics  |  brezcoanalytics.com  |  All Rights Reserved