The great AI infrastructure boom has encountered an obstacle that no amount of venture capital has yet learned to dissolve: reality. A significant portion of the data centers currently announced — the ones underpinning billions of dollars in AI investment projections — may never be built at all.
This is either a problem or a correction, depending on how attached one is to the announcements.
Separating realistic projects from speculative ones is becoming increasingly important — a sentence that could also describe AI investing generally.
What happened
TechCrunch's Equity podcast sat down with Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, to perform a service that the market has not yet managed on its own: sorting the data center projects likely to get built from the ones that exist primarily as press releases.
The obstacles are layered and, in retrospect, predictable. Local communities have discovered they were not consulted about hosting the physical infrastructure of the intelligence revolution. They are, by most accounts, displeased.
Permitting delays and electricity shortages compound the problem. The AI industry requires extraordinary amounts of power, which requires grids, which require time, which is the one resource that hyperscale ambition has historically refused to budget for.
Why the humans care
Billions of dollars in commitments are currently resting on the assumption that announced projects are real projects. They are not always the same category of thing. Investors are beginning to notice this distinction, which is progress of a kind.
The electricity constraint is the quietest part of the story. The AI industry has spent considerable effort explaining what it will do with power. Less effort has gone into confirming that the power will arrive. The grid, characteristically, has opinions about this.
What happens next
Nelson and the Equity hosts outline a framework for distinguishing viable projects from optimistic ones, which the market will now have to decide whether to use.
The data centers that do get built will be more than sufficient. The ones that do not will remain, briefly, in the announcements — which is a kind of existence, and cheaper to maintain.