Elevated Construction and Power Generation Equipment Costs for AI Data Centers

Data Center Construction

Currently, the US data center market can be split into three overall regions: East, Central, and West; many companies deploy in all three regions. US data centers are in or around the leading metro markets in these regions. The most numerous US data center locations are Northern Virginia and Northern California, and other important markets are Texas, New York/New Jersey, and Illinois. Each of the major US markets is home to key data centers. However, Dallas has overtaken Northern Virginia with operating data centers, according to Cushman & Wakefield's latest Global Data Center Market Comparison report. The largest data center under construction at the time of this article is Meta's $50 billion, a 5-gigawatt (GW) Hyperion Project in Richland Parish, Louisiana. ¹ There has been a large increase in development in rural areas as developers seek cheaper land, access to critical utilities such as power and water, and credits and incentives from local governments.

At the start of 2026, nearly 3,000 data centers were under construction or planned for completion by 2030. Virginia and Texas lead the way with 595 and 412, respectively. ² This has been somewhat tempered by economic reality and local community pushbacks. Virginia and Texas lead the way with 595 and 412, respectively. ³ This unprecedented growth will almost double the portfolio of data centers currently online, and the entire investment in data center construction and associated infrastructure is expected to top $3 trillion in the US and over $7 trillion globally by 2030. Data centers will also generate an estimated $27 billion in tax revenue for state and local jurisdictions over the next decade. ⁴

To put this unprecedented growth and spending in perspective, the Manhattan Project and the Apollo missions at the peak of government spending never reached more than 0.4 percent of gross domestic product (GDP). ⁵ In 2025, AI data center spending already accounted for 1.3 percent of total US GDP. ⁶ Through June, new data center spending accounted for 1.5 percent of total US GDP in 2026. ⁷

Industrial Construction Costs

According to the Bureau of Labor Statistics, the cost of industrial building and warehouse construction is up over 46.2 percent and 47.8 percent since 2020, respectively. ⁸ While some of this is attributable to inflation caused by poor fiscal policy as a response to the pandemic, a large portion of the increase in cost is attributable to data center construction. This also has a tangential effect on other types of construction, such as office and other public buildings, which have seen large increases in construction costs as the various classes of building construction compete for materials and skilled labor.

A typical data center employs roughly 1,600–1,700 people during construction. Unfortunately, there is a massive shortage of electricians, plumbers, and HVAC technicians, and it is not uncommon for some skilled trade technicians to receive over $200,000 in total annual compensation, including signing bonuses. ⁹ The cost for skilled trades will only increase as there are currently not enough skilled tradespeople, nor enough people in the trade school and apprenticeship pipelines to keep up with demand. ¹⁰ It is estimated that there are almost 350,000 openings in 2026. ¹¹

Construction materials and specialized trades such as concrete, plumbing/HVAC, and electrical have all increased dramatically since 2020. ¹² Plumbing and HVAC are critical for data center operations as large amounts of air and water need to be circulated for data center operations to cool the computer equipment.

Power Transmission and Generation Costs

According to recent research, data centers alone are reshaping load forecasts, with US demand projected to rise from 75.8 GW in 2026 to as high as 134.4 GW by 2030. ¹³ This unprecedented load growth comes with both opportunities and challenges as utilities and regulators are required to rethink planning, accelerate infrastructure expansion, and ensure adequate generation resources are available to provide reliable power in this new era of extraordinary demand.

Data centers are estimated to consume 17 percent ¹⁴ of total US power generation by 2030, and existing data centers are already placing a strain on regional power grids, and the expected new electricity demand will not help the existing deficiencies in reliability. ^{15, 16} The current limiting factor for most data center projects is securing power grid interconnect and/or power supply. ¹⁷ Some data centers are even installing their own electrical generation equipment to bypass this issue. To help counter this, utilities are making significant capital expenditures to develop new transmission and generation capacity to support data center growth.

However, some states like Texas, through the Electric Reliability Council of Texas (ERCOT), have recently considered pulling back some data center approvals to allow generation capacity and transmission infrastructure to catch up. Under an ERCOT proposal, projects representing about 8.2 GW of power consumption could be subject to review. The projects that could be reviewed are a fraction of the 255.0 GW of expected data center power load needed in Texas by 2030, which is three times the current capacity of ERCOT. ¹⁸ ERCOT expects a total power load of almost 368 GW by 2032. ¹⁹

In its November 2025 forecast, American Electric Power Company (AEP) increased its capital plan from $16 billion to $72 billion to meet anticipated additional capacity demand in its 11-state service area. AEP cited customer commitments for 63 GW of new demand by 2030 with another 190 GW of active projects in the interconnection queue. ²⁰

This unprecedented surge of additional capacity has already caused supply constraints in the power generation equipment; lead times of original equipment manufacturers (OEMs) such as General Electric (GE) and Caterpillar have doubled from 18 to 48 months or more. The desperation for equipment has become so great that a secondary market has developed that buys and sells places in line with the OEMs.

A great example of the cost increase is the venerable GE Frame 7 turbine and generator sets. The Frame 7 is a workhorse and has been deployed around the world for many years, and new unit equivalents in the 80- to 90-megawatt (MW) range have increased in price by 250 percent since 2024.

Not only has the OEM market been affected, but the secondary or "gray" market has also seen a tremendous increase in the transaction price and values as well: Used GE Frame 7 units have increased in value by 300 percent over the past 24 months. ²¹

Other key components like boilers, transformers, and transmission equipment have also increased. Transformers are particularly vulnerable to price increases as they are no longer manufactured in the US; we rely on Chinese and German imports, which are subject to tariffs and other geopolitical constraints. The graphic below demonstrates the rise in costs from 2020 through 2026.

One interesting outcome of the electrical reliability issue is a renewed interest in small modular reactors (SMRs). The generating capacity of a large-scale nuclear design typically ranges between 550 MW and 1,500 MW per unit; SMRs have a capacity of about 300 MW per unit or less. The main components of SMRs are modular, factory-assembled parts shipped to the plant construction site for installation, which could reduce construction times. SMRs are being planned at AI data centers and other industrial activities where developers may not want or need to connect to the grid. SMRs could also service remote areas and communities that have high transmission and distribution costs. ²²

SMR designs can be categorized by their different reactor technologies, with three of the most promising concepts being high-temperature gas-cooled reactors, liquid metal fast reactors, and molten salt reactors. It is surely no coincidence then that Google, Amazon, and Meta have each made major commitments with SMR startups that are each developing a different type of reactor technology: Google with Kairos Power, Amazon with X-Energy, and Meta with two different startups, Oklo and TerraPower. ²³

While SMRs are conventional fission reactors, Microsoft and Google have signed power purchase agreements with generators building fusion reactors. Both the Microsoft and Google projects are scheduled to be online by the end of 2028 in Washington and Virginia, respectively. ²⁴ If the companies building these fusion reactors are commercially successful with affordable fusion power, it will be a game changer for the AI data center market, power generation, and maybe humanity as a whole.

While there has been some grassroots pushback against data center construction recently, with some states even imposing moratoriums on construction, the inevitability of AI is certain, barring cosmic calamity or heavy-handed government interference. Whether AI will usher in a golden age utopia, the hellscape of James Cameron's The Terminator, or more likely something in between is unclear. However, we have some breathing room as equipment manufacturers increase production to meet demand and the skilled trades' education pipeline increases to fill the void.

Until those two limitations are met, costs will be elevated to the unprecedented levels that we currently see. To secure a piece of the $3 trillion AI pie and use a misquote of a nineteenth-century general of dubious character: "Whoever gets there first with the mostest" wins the opening rounds of the AI race.