U.S. Electricity Under Pressure: AI Energy Demand
- 3 hours ago
- 7 min read
The growing electricity demand fueled by AI and data centers presents significant challenges to the energy landscape. However, U.S. power generation, which has grown only slowly over the past decades, is struggling to keep pace with this surge. This blog also highlights issues related to power generation and the aging U.S. grid, which struggle to keep pace with the need for reliable electricity. Considering these challenges, tech companies are exploring innovative strategies, including developing their own power solutions and securing long-term contracts, fundamentally transforming how energy is generated and distributed.
Why This Matters
If electricity demand from AI and data centers skyrockets, power shortages could disrupt the tech industry and everyday life.
If current energy sources can’t keep up, economic growth driven by AI will face serious limitations inhibiting a key strategic growth engine for the U.S.
If the challenges of the power grid are ignored, issues of rising electricity demand will lead to soaring prices and potential fragmentation of the power grid.
If companies don’t proactively secure their energy sources, they will find themselves at a disadvantage in the AI competition.
AI and Data Center Electrical Shock
AI is booming. Changes in annual energy consumption in 2023 reveal that, while the median energy consumption of S&P 500 companies remained flat, it rose significantly for the “Magnificent Seven” (the seven largest US tech companies whose development are linked to the rising use of AI) and data center firms, by 19% and 7% respectively. In practice, each ChatGPT question is estimated to use around 10 times more electricity than a traditional Google search.
In 2024, data centers consumed about 415 terawatt hours (TWh) of electricity worldwide, accounting for roughly 1.5% of global electricity consumption. Over the past five years, this demand has grown at an average annual rate of around 12%. Looking ahead, global electricity consumption from data centers is projected to more than double to about 945 TWh by 2030, approaching 3% of total global electricity use. Between 2024 and 2030, data center electricity demand is expected to grow at around 15% per year, more than four times faster than electricity consumption growth across all other sectors combined. Driven by the rapid expansion of AI and data centers planned by 2035, global demand for electricity is expected to increase by more than 10,000 TWH, [LD1] roughly a one-third increase relative to today’s global consumption and equivalent to the total consumption of all advanced economies today.
The projected surge in electricity demand from data centers is even more pronounced in the United States. According to International Energy Agency (IEA) estimates, U.S. data centers consumed 183 terawatt-hours (TWh) of electricity in 2024, accounting for more than 4% of total national electricity consumption. That level of demand is roughly equivalent to the entire annual electricity use of Pakistan. Looking ahead, U.S. data center electricity consumption is projected to grow by 133% by 2030, reaching around 426 TWh.
This demand is not evenly distributed across the country. According to Data Center Map, the United States hosts more than 4,000 data centers, including both operational facilities and sites under development. Nearly one-third of these facilities are concentrated in just three states, namely Virginia, Texas, and California. Such geographic concentration magnifies local grid stress. In 2023, data centers in Viginia accounted for approximately 26% of total state electricity consumption. Similarly, large shares were seen also in North Dakota (15%), Nebraska (12%), Iowa (11%), and Oregon (11%), according to the Electric Power Research Institute (EPRI).
In contrast to surging demand, electricity supply in the U.S. has remained sluggish. While the share of clean energy in the generation mix has steadily increased, overall electricity output has remained largely flat. This stagnation leaves the power system poorly positioned to absorb the sharp increase in generation demand driven by the rapid expansion of the AI industry.
U.S. Grid Mismatch
Even if sufficient generation capacity were available, the U.S. power system would still struggle to absorb electricity at the scale required, due to the fragility of its transmission and distribution network.
First: The first constraint lies in the regional nature of grid governance: the U.S. does not operate electricity transmission under a single, unified system with one set of rules. Power transmission is instead coordinated by regional Independent System Operators (ISOs) and Regional Transmission Organizations (RTOs), each with its own market design and planning priorities. In practice, the U.S. is commonly described as having seven major ISO/RTO regions (PJM, MISO, CAISO, SPP, NYISO, ISO-NE, ERCOT).
These ISO/RTO regions account for roughly two-thirds of U.S. electricity load. This fragmentation means that even when generation capacity looks sufficient at the national level, electricity may not be deliverable at the right location, at the right time, or under the right operational and market conditions. For data centers whose demand is highly concentrated and requires firm, predictable supply, the mismatch caused by regional market boundaries can be further amplified.
Second: A second constraint lies in the difficulty of building new transmission infrastructure. In the United States, building new transmission lines, especially long-distance and interstate ones, requires navigating multiple layers of approval, including state and local permitting, route siting, environmental reviews, and negotiations with numerous landowners over rights-of-way. Opposition at any stage can delay projects for years or trigger litigation.
Reports show that transmission siting and permitting in the U.S. has been recognized as a long-lasting and structural issue, with persistent challenges arising from land-use and community resistance.
Third: And finally, limited capacity and age of the existing grid act as the third constraint. From the 1970s through the 2010s, U.S. transmission infrastructure expanded at an average rate of about only 1.5% per year, while electricity demand largely plateaued. As a result, much of today’s grid was built in the 1960s and 1970s and has now been operating for more than half a century. According to data from the U.S. Department of Energy, nearly 70% of grid assets are more than 25 years old, with many transmission lines and transformers operating well beyond their original design life. These legacy constraints have become increasingly binding in the AI era, yet upgrading the grid at the required scale remains slow and institutionally difficult.
Tech Companies and Power
Tech companies in the AI race need power, and they aren’t relying on the archaic U.S. power grid to catch up. They are actively reshaping how power is generated and distributed.
One approach is to build power generation directly next to data centers. For example, in West Texas, OpenAI and Oracle’s massive $500 billion Stargate project involves plans to pair hyperscale compute with on-site natural gas generation. This strategy is increasingly supported by regulatory changes. As an example, in Oklahoma, state legislation now explicitly allows businesses the to develop their own power infrastructure.
Another approach is directly locking in power supply from existing generation assets. In Pennsylvania, Amazon has secured long-term electricity arrangements linked to the Susquehanna nuclear power plant to support its data-center operations. Similarly, Microsoft has signed a 20-year agreement with Constellation Energy, the largest clean power producer in the U.S., to allow Unit One of the Three Mile Island nuclear power plant in Pennsylvania, to be restarted and anchor its data-center growth to dedicated nuclear output. In both cases, electricity remains physically on the grid, but from an economic perspective, capacity is increasingly contractually tied to hyperscale customers rather than allocated through the shared market.
In the short term, some companies have turned to on-site gas turbines as an emergency measure to meet immediate power needs. In Memphis, Tennessee, at one point, Elon Musk’s xAI has reportedly relied on up to 35 methane-fired gas turbines to power its data-center operations. However, the use of gas turbines is widely understood as a temporary stopgap rather than a durable solution. Gas turbines represent a capital-constrained segment of the power industry, where manufacturing capacity has lagged demand after years of underinvestment, resulting in extremely tight supply. Beyond scalability limits, on-site turbine deployment has also raised concerns over local air pollution and legal and regulatory challenges.
Looking further ahead, another strategy is to invest directly in energy startups to secure future power supply rather than waiting for utilities or markets to respond. Sam Altman, the CEO of OpenAI, has backed Oklo, a nuclear energy company developing small, advanced fission reactors to power data centers and other high-demand energy users. Similarly, Google has backed Fervo Energy, a company focused on next-generation geothermal power. Google began partnering with Fervo as early as 2021 on the world’s first corporate agreement to develop an enhanced geothermal power project. In December 2025, Google participated in Fervo’s $462 million funding round, underscoring a longer-term effort to scale firm, carbon-free power.
A shift away from the traditional model of simply paying for grid electricity toward generating power for on-site use first emerged as a reluctant choice by data-center operators facing grid constraints. Today, that shift is increasingly being formalized by governments. In December 2025, the Irish government published Large Energy Users Connection Policy. The decision paper confirms that new data centers planned for Ireland must have dispatchable generators or batteries, either on-site or nearby, with capacity matching their grid import limits. Electricity generated at data center sites would be fed back into Ireland’s electricity system through the wholesale market. In addition, operators are required to source 80% of their annual electricity demand from renewable sources, with a six-year glide path to transition into compliance. This policy framework could serve as a valuable reference for the U.S. market, offering a potential model for integrating large energy users into grid planning while advancing renewable energy goals.
Take Action
Policymakers & Regulators: Develop policies to facilitate faster permitting and deployment of new energy infrastructure, ensuring that the grid can meet the rising demands of AI and data centers.
Utility Providers: Collaborate with tech firms to innovate solutions that integrate clean energy sources into the grid, ensuring reliable supply while supporting sustainable practices.
If you’re interested in exploring similar topics and ideas on ClimateTech and Clean Energy Solutions, then please check out the The Wall Street Green Summit happening 10th & 11th March, 2026, please register here.
