AI Infrastructure Bids Farewell to the Concept Buying Era, Physical Constraints Spurring 'Electricity Premium'
Taylor Wilson
North American data-center vacancy has dropped below 2% and grid interconnection queues stretch past six years — grid capacity, not hardware orders, now sets the price of AI infrastructure.
Where exactly is the bottleneck?
North American core-metro data-center vacancy has fallen below 2%, and grid interconnection queues now commonly exceed six years.
Goldman Sachs counts 48 data-center projects delayed in 2025 by local opposition over power pricing and water use, involving $156 billion in capital.
This means → even unlimited funding cannot jump the queue — grid access and permits are the real hard constraint.
Who has to prove the power bill is worth it?
Investors no longer judge cloud giants by how many GPUs they buy; the new metric is how much real cash flow each gigawatt of power capacity generates.
In plain terms = cloud operators must cover massive depreciation bills with actual revenue — public-cloud sales, ad-recommendation efficiency gains, and the like.
Among the majors, Microsoft's cloud-plus-enterprise-software synergy makes its monetization path clearest; Google holds a vertical-integration edge; Meta, lacking a public-cloud buffer, faces the steepest pressure to prove returns.
How has the asset pecking order changed?
The top tier is no longer "the company with the best AI story" — it is utilities with secured grid positions, transformers, and distribution infrastructure, plus base-layer data centers with locked delivery dates.
The second tier consists of compute projects backed by big-tech long-term leases or hard-commitment contracts — predictable enough to be securitized, which lowers their cost of capital.
This reflects a fundamental shift: the market's pricing anchor has moved from growth narrative to physical delivery capability.
Which assets face a reckoning?
Morgan Stanley notes that even after accounting for new energy solutions, U.S. data-center demand will significantly outstrip available grid capacity through 2028.
This means → interconnection and construction risk cannot be solved by capital alone — pure-narrative assets with no physical delivery capability face a washout.
Put simply = the market used to buy "who tells the best AI story"; now it buys "whose power arrives first."
Content is for reference only, not financial advice.