Construction has continued in a section of northern Virginia known colloquially as “Data Center Alley,” which runs along the Route 28 corridor through Loudoun County and into Prince William County. Existing facilities are being replaced with new ones. There is an expansion of power substations. Instead of office parks, the traffic on the roadways close to some locations is more like that of large-scale industrial operations.
In 2023, data centers in Virginia used 26% of the state’s electricity. The number continues to grow, and both engineers and regulators believe that the grid that supplies these facilities was not designed to handle the demands placed on it. Beneath a trillion-dollar investing premise is this actual reality.
It is anticipated that Alphabet, Amazon, Microsoft, and Meta would increase their data center infrastructure spending from $350 billion in 2025 to almost $400 billion in 2026. According to Morgan Stanley Research, prominent technology businesses have committed more than $1 trillion in total for the 2025–2026 period alone. Even those who follow capital markets professionally often get a sense of vertigo when they see these statistics because they are too big to completely comprehend. However, the underlying constraint is not simply financial. It’s tangible.
When a new data center files a request for grid connections in some U.S. markets, it may have to wait seven years to receive power. As hyperscalers fought for the same available generation capacity in 2024, the cost of power purchase agreements increased by 35%. According to Morgan Stanley, the demand for data center electricity in the United States might reach 74 gigawatts by 2028, with a 49 gigawatt shortage. The money is available to construct the infrastructure. The electrons are more difficult.
If anything, the ROI question is more difficult to answer in the near future than the energy question. Over the past two years, CFOs from a variety of businesses have watched the deployment of large AI infrastructure with increasing impatience for quantifiable outcomes. In certain applications, such as software development acceleration, customer service automation, and some types of document processing, the technology has actually increased productivity.
However, for many organizations, the direct revenue return that would support capital deployment at this scale is still somewhere ahead of them rather than in the current quarter’s numbers. This does not prove that the investment is flawed. It’s a timing observation. Before the profits materialized, every significant infrastructure project in the history of technology appeared dubious at the point of maximal capital investment. Whether the curve bends in the anticipated direction on the anticipated timetable is the question.
In addition to the operational difficulties, the market concentration component introduces a structural risk. While smaller AI startups trade at revenue multiples that many analysts quietly describe as difficult to justify on current fundamentals, venture and corporate investment has heavily concentrated in a small number of U.S.-based hyperscalers and frontier AI developers, including OpenAI, Anthropic, Google DeepMind, Microsoft, and a few others. Data centers accounted for 21% of Ireland’s total power consumption in 2022, and projections indicate that this percentage will rise to 32% by 2026.
This raises the kind of political and community opposition that doesn’t appear on a spreadsheet until a permission is delayed or a site relocation is required. According to Microsoft’s own 2025 sustainability report, there has been a 23.4 percent increase in overall emissions and a 168 percent increase in energy usage since 2020, both of which are directly related to the growth of AI and cloud computing. Regulators are beginning to pay closer attention to these externalities.
It seems as though the AI industry has reached a point when the initial idea and the available infrastructure are being put to the test at the same time, given the magnitude of the financial commitment and the physical constraints. Given the size of the investment wager, it is likely to discover solutions, such as on-site power systems, modular nuclear, or increased gas output. The seven-year grid queue, however, is not a forecast. It is a fact of the present. Additionally, most organizations continue to rely on rather than confirm the returns that would make all of this clear in retrospect.
