Data center executives say power and labor, not chips, are now the AI buildout's limiting factor

At Bloomberg's New Economy Forum in New Delhi on October 13, data center executives from NTT Data, Brookfield Asset Management and Infosys described a shift in what is actually slowing the AI infrastructure buildout. For much of the past two years, the industry's bottleneck narrative centered on GPU supply. That framing no longer holds, the executives said: electricity access and a shortage of skilled construction and electrical labor have become the binding constraints on how fast new AI data centers can be built.

NTT Data chief executive Abhijit Dubey called skilled labor "the bottleneck nobody talks about." He said pay for a qualified U.S. electrician has risen to roughly $200,000 a year as hyperscalers and developers compete for a limited pool of electrical, design and construction talent, and that sourcing the right equipment for a build can now take up to 42 weeks. Dubey said NTT Data has capital available to expand its AI data center footprint but cannot deploy it fast enough because of these constraints — directly at odds with the idea that financing, rather than physical build capacity, is what limits the industry.

Brookfield: demand will outstrip supply "past the end of the decade"

Brookfield Asset Management's Connor Teskey, who oversees the firm's AI infrastructure investments, said the real test for the industry is operational rather than financial. Execution, he argued, is now the defining differentiator among developers.

The bottleneck is who can pull these projects out of the ground.

Teskey rejected the argument that AI infrastructure spending constitutes a bubble, noting that the large majority of projects under construction are backed by long-term, take-or-pay contracts with investment-grade hyperscaler tenants — a structure he said locks in demand regardless of near-term swings in model economics. His framing was not unanimous at the forum: a day earlier, Broadcom's chief executive argued that capital remains the industry's biggest constraint, underscoring how differently chip suppliers and infrastructure operators are reading the same buildout.

Permitting and public pushback compound the physical constraints

Infosys chief executive Salil Parekh named a third constraint: public opposition. Grassroots and local-government resistance blocked roughly $130 billion of U.S. data center projects in the first quarter of 2026 alone, and pushback has forced reversals elsewhere — Dublin rejected a proposed Google data center over energy concerns, and Scotland's governing party voted to freeze new hyperscale approvals, affecting 24 projects whose combined power draw would exceed one and a half times the country's peak demand.

Taken together, the comments mark a notable change in how the industry's most capital-intensive buyers describe their own limits. Bloomberg has separately put the cost of powering a single large U.S. data center at up to $3 billion. Combined with multi-year grid-connection queues and a tightening skilled-trades market, that suggests the AI buildout's pace through 2027 will be set less by chip shipment schedules than by utilities, electricians and local planning boards.