The race to build AI infrastructure is hitting a hard wall: the United States simply cannot deliver electricity fast enough to power the data centers being constructed. While companies like Microsoft and OpenAI are pouring $1.7 trillion into data center expansions between 2024 and 2030, the physical infrastructure needed to run these facilities is lagging years behind. The result is a widening "data center development gap" that threatens to slow the AI revolution itself.

Why Is the Power Grid Becoming the Real Bottleneck?

The problem starts with sheer scale. A single modern AI data center campus can demand over 1 gigawatt (GW) of power, equivalent to the electricity needs of a small city. OpenAI's Stargate campuses in Texas and Michigan will each exceed 1 GW once fully operational. This is a dramatic shift from pre-AI data centers, which typically consumed far less power.

The interconnection queue tells the story. As of the end of 2025, over 2,060 GW of total generation and storage capacity were seeking connection to the U.S. electrical grid. The United States has only about 1,250 GW of utility-scale electricity generation capacity available. That means current interconnection requests equal 165 percent of available capacity. In other words, the grid is being asked to do something it physically cannot do right now.

Developers in Northern Virginia, Arizona, Texas, Georgia, and Ohio increasingly face multi-year timelines just to secure reliable power delivery. Utility providers are requiring extensive substation upgrades, transmission expansion, or entirely new generation capacity before projects can proceed. Initial utility interconnection now takes 5 to 7 years, while new transmission upgrades take 5 to 10 years or more. By contrast, data center construction from groundbreaking to operational readiness typically takes 18 to 36 months for standard hyperscaler facilities, or 36 to 60 months for AI mega-campuses. Without preexisting interconnection agreements, developers can build data centers faster than power can be delivered to them.

What Supply Chain Problems Are Slowing Construction?

Even operators with secured utility agreements face severe delays tied to global supply chain disruption. The breakneck pace of data center demand has created shortages for transformers, switchgear, generators, cooling systems, and high-voltage electrical equipment. Procurement timelines have stretched from months to years as global demand outpaces manufacturing capacity.

Power transformers, which convert high-voltage grid electricity into lower-voltage, usable energy, are the clearest example. Modern AI data centers require multiple utility-scale power transformers on each site. Demand for transformers has spiked 116 percent, and costs per unit have increased 77 percent since 2019. The United States imports 80 percent of power transformers and 50 percent of distribution transformers from overseas manufacturers, primarily in China, Mexico, South Korea, and Japan.

Gas turbines, increasingly essential as developers turn to behind-the-meter generation to bypass interconnection queues, have become another critical pressure point. Global gas turbine orders reached 110 GW at the end of 2025, but global manufacturing capacity is only 60 to 70 GW. This imbalance has pushed prices up an estimated 195 percent since 2019 and stretched lead times to roughly six years. U.S. tariffs and geopolitical "de-risking" efforts have further increased costs, with neither reindustrialization nor procurement diversification addressing the near-term supply crunch.

How Are Communities and Governments Pushing Back?

Historically, state and local governments welcomed data centers for the tax revenue and economic development they promised. Loudoun County, Virginia, nicknamed "Data Center Alley," became a model for this approach. Starting in 2010 when Virginia enacted a landmark data center sales-and-use tax exemption program, the county's annual budget grew from about $1.7 billion in 2010 to over $5.4 billion in 2027. In the FY2027 budget, data centers alone account for nearly $1.3 billion, or roughly 45 percent of tax revenue.

But that political consensus is cracking. Since the start of the 2025-2026 legislative session, twelve state legislatures have proposed some form of statewide moratorium on data centers. Maine's legislature was the first to pass a statewide moratorium, pausing development of data centers exceeding 20 megawatts (MW) of capacity until late 2027, though Governor Janet Mills subsequently vetoed the measure. Several governors and state legislatures have also proposed altering existing tax exemptions and subsidy policies for data centers, including Virginia, Illinois, and Georgia. Washington became the first state to roll back certain data center tax exemptions when Governor Bob Ferguson signed Senate Bill 6231 into law on April 1, 2026, cutting a sales tax exemption for data center operators replacing or refurbishing old server equipment.

Growing scrutiny of employment benefits is also fueling the shift. While developers have long touted new projects as engines for job creation, many employment opportunities come in the form of temporary construction roles that disappear once a facility goes online. A 2024 study by the Virginia Joint Legislative Audit and Review Commission found that a typical 250,000-square-foot data center creates far fewer permanent jobs than initially promised.

Steps to Address the Data Center Infrastructure Crisis

• Accelerate Interconnection Timelines: Utilities and regulators must streamline the interconnection approval process, which currently takes 5 to 7 years. Parallel processing of applications and pre-approved transmission corridors could reduce delays significantly.
• Diversify Supply Chains: The U.S. must reduce dependence on overseas transformer and turbine manufacturers. Domestic reindustrialization and procurement diversification can address the near-term supply crunch and reduce vulnerability to geopolitical disruptions.
• Implement Behind-the-Meter Generation: Data centers can generate their own power on-site through solar, batteries, gas turbines, or potentially small modular nuclear reactors, reducing strain on the public grid and enabling faster project deployment.
• Design Flexible Demand Systems: Data centers capable of ramping electricity usage up or down in response to grid conditions can stabilize the grid rather than destabilize it, becoming partners to utilities instead of problems.

What Does This Mean for AI Infrastructure Investment?

The data center development gap represents a critical inflection point. According to Bloomberg data cited in the analysis, of the 12 GW of data center capacity set to come online in 2026, only 5 GW are under active construction. This trend extends to 2027, where merely 6.3 GW of the 21.5 GW of announced capacity has broken ground.

The underlying issue is a broad phase of market reconciliation in which infrastructure development must align with the practical limitations of the domestic grid, supply chains, political institutions, and community tolerance. Projects underwritten on optimistic assumptions are seeing operational delays ranging from quarters to years, or outright cancellations.

Beyond electricity, water consumption is equally critical but often overlooked. A modest 1 MW data center using traditional evaporative cooling can consume roughly 20 to 30 million liters of water per year. At the hyperscale end, a single data center campus in Iowa consumed approximately 3.8 billion liters of water in 2024 alone. Studies estimate that around 70 to 80 percent of cooling water drawn by traditional systems is lost to evaporation. Modern facilities increasingly rely on closed-loop liquid cooling, air-based systems, and site selection in cooler climates to reduce water intensity. Microsoft, for example, has committed to eliminating evaporative water use in its next-generation AI data centers, targeting a roughly 95 percent reduction in cooling-related water consumption.

The dominant narrative frames AI data centers as threats to electricity grids. But a flexibly designed data center can actually stabilize a grid. "Behind-the-meter" power means a data center generates some of its own power on-site through solar, batteries, gas turbines, or potentially small modular nuclear reactors. "Flex power" or "flexible demand" means designing facilities that can ramp electricity usage up or down in response to grid and electricity supply conditions. When national wind and solar power plants generate excess supply, data centers can absorb that surplus. When the grid is stressed, they can curtail load.

"Utilities increasingly need this kind of productive flexible demand and data centres capable of offering flexibility can become partners to grid operators, not problems," noted Sinenhlanhla Zulu, Investment Principal at African Infrastructure Investment Managers.

Sinenhlanhla Zulu, Investment Principal, African Infrastructure Investment Managers

The infrastructure choices being made today about where data centers go, who owns them, and under what conditions they operate will shape the economy, sovereignty, and opportunity for the next 30 to 40 years. These decisions will be difficult to undo once made. The window to shape those outcomes rather than inherit them is open but will not stay open indefinitely.