Bloomberg Surveillance · Thursday, September 24, 2026
Jonathan Maxwell identifies significant challenges for powering AI infrastructure, including average grid connection wait times of 5.2 years in the U.S. and Europe. He also points out substantial energy loss as heat, estimating 53% in the UK and 65% in the U.S. from the energy source to its point of use, hindering overall efficiency.
“Well, one of the problems to get before the solution is there's a queue, right? I mean, they've got 5.2 years on average. You have to wait for a grid connection in the U.S., roughly the same five years in Europe. So the first thing is, what do you do with the queue? The second thing is, when you get the energy to the data center, how much has been wasted on the way? You know, drama, I thump the whole time. But roughly speaking, in the UK, it's 53%. In America, it's 65%. So every time you come in here, that blows me away. From the source of the energy to where it's used, you lose about half of it.”
“So most of it's lost as heat. So if you put energy in a generator, whether it's gas or uranium for nukes, what happens is the energy goes in, the primary energy, the fuel goes in, whatever it is, and then You produce two products, power and then heat. Now, if you don't do anything with the heat, it goes up into space and they don't need it up there. So that's the biggest loss in the world, which is heat loss. There's transmission distribution losses. But to answer your question, if you want to do a data center well, A, you interact with the grid and do the deal that you can. But good luck because you're going to wait five, ten years in the U.S. But B. Bring the energy, make the energy where you need it. And then you can take the heat off, use the cools, use the heat that comes off for industrial processes. And it doubles the efficiency.”