The top four tech companies have committed roughly $700 billion to AI data center construction in 2026 alone. But to become AI infrastructure, that money needs traditional infrastructure — power, water, and permits — and now Big Tech is discovering, the hard way, that America can’t deliver physical infrastructure on schedule.
Gartner projects that 40% of AI data centers will be constrained by power availability by 2027. Between 30% and 50% of facilities scheduled to open this year are already stalling. In many regions, connecting to the grid now comes with a wait measured in years, not months.
Water systems that were already strained are being asked to cool thousands of new server racks, and permitting timelines stretch so far past capital deployment schedules that investors are watching their cost of carry climb while shovels sit idle.
REGULATORS ORDER OVERHAUL OF TRANSMISSION RULES TO SUPPORT AI AND DATA CENTER BOOM
It may seem like the tech industry has triggered something new. It has not.
The oldest pattern in American infrastructure
The United States has been running this same playbook for decades: infrastructure need identified, commitments announced — and then the project dies slowly inside environmental reviews, permitting fights, funding paralysis, and political indecision while costs compound.
McKinsey estimates roughly $1.5 trillion in proposed infrastructure projects are currently trapped in permitting bottlenecks, producing annual economic losses of $100 billion to $150 billion. Construction costs have risen by 70% since 2020, which means every year of deferral costs more than the previous year.
That permitting slow-walk alone costs thousands of dollars per household annually — a tax that never appears on anyone’s bill but gets paid just the same.
The American Society of Civil Engineers gave U.S. infrastructure a “C” in its 2025 report card — the best grade since 1998 — and still projects a $3.7 trillion investment gap that is growing, not shrinking. The pattern is identical across every sector: a known need and an institutional failure to decide.
Bridges, broadband, and server farms
The infrastructure stumbles are everywhere. The $42 billion BEAD rural broadband program was allocated in 2021 to connect communities that have waited years for reliable high-speed internet access. Five years later, implementation is only now beginning after rounds of delays and rule changes. The money was committed. The demand was already there. Thousands of towns are still waiting.
This is the trajectory AI data centers are now on. A facility that pencils out today at current power costs and construction prices will not pencil out after three years of permitting review and grid interconnection delays. Viable projects become federal dependency cases through the slow accumulation of deferred decisions that compound until the original economics are unrecognizable.
Fear, not regulation
The instinct is to blame regulation, but that misses the target. The deeper problem, though, may be political fear. Officials at every level have learned that announcing a project and delivering one are not the same thing — and that the gap between the two rarely costs anyone their job.
Ordering a study carries fewer political risks than delivering a result. Saying yes to a data center means owning the trade-off on water, power, noise, and land use. Saying “we need more review” doesn’t.
That isn’t a partisan observation, but a structural one. Leaders in red states and blue states alike have learned that delay is the lowest-risk political strategy, even when it is the highest-cost economic one.
The old infrastructure crisis can take us down
If these severe self-inflicted bottlenecks persist, the consequences will extend well beyond delayed AI infrastructure. The delays have the potential to reorder where artificial intelligence gets built and, ultimately, who leads it.
If power, water, and permitting cannot be secured in the U.S. on predictable timelines, AI investment will flow to jurisdictions that can deliver them. That means data centers, training clusters, and the next generation of AI infrastructure will increasingly be sited in countries willing to make faster decisions, even if those decisions come with different regulatory, security, or governance trade-offs.
The long-term effect will be — not might be — a quiet erosion of the enormous American advantage in AI. Leadership in this sector is not determined solely by model innovation, but by the ability to deploy and scale it. When infrastructure becomes the constraint, the locus of innovation will follow the locus of deployment, just like the current flight of capital from blue to red states.
If the U.S. can’t translate financial commitment into physical capacity, we risk becoming a designer of AI systems that are trained, operated, and monetized elsewhere. That, in turn, would represent a failure of government’s ability to deliver the next generation of public services in a fundamental and lasting way.
Accountability is the only fix
The answer is not more funding announcements or ribbon-cutting photo ops. The answer is accountability — treating timelines as binding commitments and attaching decision-making authority to decision-making responsibility.
Case in point: The largest private infrastructure project in Texas history, Dallas’s LBJ Express, opened three months ahead of schedule because private investors owned the deadline and bore the cost of missing it. When delay stops being an abstraction and becomes a number on a balance sheet with a name next to it, projects get built.
PROTECTING CONSUMER GRID AND CORPORATE RESPONSIBILITY AREN’T AT ODDS
The public sector needs a version of the same mechanism: fixed decision windows with automatic escalation, personal accountability for officials who hold authority over timelines, and consequences — political or financial — for delay without cause. No bridges, broadband gaps, or data centers will be built simply by announcing them and posing with a giant check.
The U.S. can only maintain its position of technological superiority in the age of AI by providing functional public infrastructure and by clearing the approval tracks for the building of private AI infrastructure. If our public officials fail at that task, our private AI companies are likely to fail at theirs.
Bob Hellman is CEO of American Infrastructure Partners, a private investment firm focused on U.S. infrastructure.