AI's Opportunity In Building Data Centers

Shiva Dhawan is the CEO and cofounder of Attentive.ai, the company behind Beam AI.

There’s a lot of debate about AI bubbles. Most of it focuses on valuations, model economics or whether hyperscalers can earn a return on the hundreds of billions they’re spending on AI infrastructure.

However, a more immediate mismatch separates imagined value from reality. We’ve become great at financing data centers and much less capable of actually building them.

J.P. Morgan’s 2026 Eye on the Market report noted that of the roughly 47 gigawatts of U.S. data center capacity expected to be worked on in 2027, satellite data showed no observed construction activity on more than half. Permitting delays, turbine and transformer shortages, and skilled labor gaps are keeping projects from moving from plan to reality.

At the same time, the money pouring into AI infrastructure is enormous, and it’s not stopping.

A 2025 McKinsey & Company report estimated that meeting global data center demand could require $6.7 trillion of capital expenditures through 2030. A J.P. Morgan report on AI predicted hyperscaler capital expenditure would consume roughly 90% to 100% of operating cash flow in 2026, with companies increasingly supplementing that spending with debt and private credit. Nvidia announced in August 2026 that more than $500 billion in third-party capital would go toward hyperscalers, AI labs and data center build-outs.

This doesn’t mean the AI boom is imaginary. Demand for compute is real, existing data centers are tight, and J.P. Morgan’s AI report recorded a vacancy of just 1.4% in major U.S. markets. The problem is that an announced gigawatt is different from an operating gigawatt.

The constraint is increasingly physical.

Wall Street can move money at digital speed. Construction can’t. A modern data center requires land, permitting, power generation, enormous quantities of electrical equipment and thousands of people to put it all together.

Labor is only one part of the problem.