Key Takeaways

  • Industry estimates show roughly 50% of announced AI data centers will never come online due to severe execution roadblocks.
  • Large load interconnection agreements with local utilities and air permits for new power generation create multi-year approval delays.
  • Physical real estate deals often stall when landowners refuse to sell or local entitlement approvals face community resistance.
  • The United States faces a hard ceiling on skilled trade workers, leaving projects stalled without enough electricians, welders, and plumbers.

The Real Attrition Rate of AI Infrastructure

Announcing a multi-gigawatt data center campus is easy. Breaking ground and energizing thousands of high-density server racks is entirely different. While venture capital pours billions into AI compute, a massive portion of planned capacity will never materialize.

Chase Lochmiller, co-founder and CEO of Crusoe (which recently closed a $3.9 billion Series F), sees this friction up close. When asked about projections that half of planned facilities will fail, Lochmiller agreed: “We've seen very prominent people say, 'Oh, we expect 50% of data centers that are planned not actually to be built and ready to go.' What percent of data centers do you think that are planned will not actually go online? 50% seems reasonable.”

Building physical infrastructure in Western democracies requires dealing with hundreds of independent stakeholders. Unlike state-directed models that clear land by decree, American projects face local zoning boards, environmental reviews, and property negotiations. Lochmiller compares the process to high-altitude climbing: “It's one of these things where, you know, kind of coming back to this notion of thinking like a mountaineer when you're going through these planning processes. A lot of things can go wrong.”

Power Interconnections and Land Friction

The earliest point of failure is often the physical site itself. Developers map out vast tracts of land, only to run into uncooperative property owners and zoning barriers. Lochmiller points to the friction of “getting permits, getting entitlements, getting land... if there's, you know, you're trying to acquire a certain piece of land and, you know, the person you're trying to buy it from ultimately doesn't want to sell.”

Even with secured land, power remains the ultimate gatekeeper. Drawing hundreds of megawatts from a regional grid requires complex commercial and technical approvals. Lochmiller emphasizes the challenge of “getting a large load interconnection agreement done with the local utility” as well as “getting an air permit if you're bringing online new generation.” Utilities take years to study grid impacts, upgrade substations, and approve transmission links. Without those signed agreements, a project is merely an architectural drawing.

The Skilled Labor Ceiling

Money cannot instantly create human expertise. Even when a developer secures land, power, and environmental permits, construction runs directly into a severe domestic labor bottleneck. AI facilities demand specialized industrial trades to assemble liquid-cooling loops, high-voltage transformers, and complex backup power systems.

“Energy is definitely a key constraint so power that's available for compute is a key bottleneck,” Lochmiller explains. “Alongside that, labor is a very key bottleneck. There are a finite amount of skilled trade workers: electricians, welders, plumbers, construction workers in the United States.”

Because every hyperscaler and infrastructure startup is competing for the same limited pool of union halls and certified contractors, project timelines stretch and labor costs escalate. When contractors fail to show up, schedules slip by quarters or years.

What to Do With This

If your startup relies on future GPU capacity commitments, audit your cloud providers today. Ask your vendors whether their upcoming capacity is backed by signed utility interconnection agreements and active construction permits, or if it sits in early development. If they cannot prove power rights and energized delivery dates, assume a 50% discount on their projected capacity timelines and diversify your provider base immediately.