Mark Moss breaks down the infrastructure gap behind AI expansion. U.S. utilities plan to spend $1.3 to $1.4 trillion on power and cooling between 2026 and 2030, but that's what they've already budgeted, not what AI actually needs. The real buildout will require far more capital than utilities currently have lined up.
So the problem with power and cooling is that somebody has to build it still. All right, now we can see that US utilities are planning to do that. They have about $1.3 to $1.4 trillion of capital expenditures, what we call CapEx, planned between just 2026 and 2030. We can see Edison Electric, S&P Global Market Intelligence, and Powerlines all come out with about the same roughly number. Now that's planned spending. Okay, that's not an estimate of what's needed. Nobody separates out how much of that is maybe in the data centers. But the thing that unlocks the abundance is itself an enormous call on new money that needs to be created to do that.
