The case for colocating data centers and generation

Kimber starts from a premise he has been making for years: the grid is broken, so load has to come to generation rather than the other way round. Three trends are pulling load up at once, and he is careful that only one of them is about climate. Digitalization is the loud one. Electrification is the one he finds more interesting, because it is increasingly happening on product merit rather than policy, as electric motors and arc-furnace steelmaking simply get better. Decarbonization is the third and, in his view, survives whatever the policy environment. Against that, the conventional path is to secure a site and wait roughly a decade for a transmission line. He grants that works if you have the patience and the capital, and that there have been successes, but he doubts it is the solution set the new loads need.

The case for colocation is really three separate claims of three different kinds, and colocation arguments routinely blur them. The first is regulatory and is about who pays. In most jurisdictions, he says, the network upgrades triggered by a large load interconnection get socialized across ratepayers, so a hyperscaler’s transmission line ends up on everyone’s bill, whereas a generator interconnection usually makes the generator cover more of the substation and related work. Shifting the interconnection from the load side to the generation side changes who bears the cost, and that is a rules question rather than a physics one. The second claim is engineering, and turns on the difference between energy and capacity. Most places, he says, have enough energy in most hours; the real question is whether the new load pushes the system over at its peak moments. Onsite batteries, dispatchable gas or the solar itself can keep the site from contributing at those moments. The third is about permission: renewables are still the fastest generation to bring online in the United States and will be for at least another five or six years, and community and regulatory acceptance is already a constraint on renewables, which he expects to bite harder on new combined-cycle plants and reactors.

The economics then support a reframe of what baseload means. In the Texas panhandle, where Intersect’s largest sites sit, renewable capacity factors reach the high seventies, so wind, solar and batteries on one site can run the load fully carbon free for 75 to 80 percent of the hours in a year, with a bit over 20 percent covered by gas or by the grid. The gas he means is simple cycle or reciprocating engines, not combined cycle, which he dismisses for this purpose. He claims that package beats a new combined-cycle plant on cost, given the price he quotes for one, a five to six year build, and a construction labor pool that in many cases has never built one. Note what the claim is and is not: he is not saying gas is unnecessary, he is saying the gas actually needed is a different, smaller and faster kind than the debate assumes.

The limit on all of this is the demand side, and here he is blunt. Asked whether data centers will run flexibly, he says power is still the tail of the dog and is not wagging anything. His rough capital cost stack for a gigawatt site is what carries the argument: on the order of ten billion dollars of data center structure, five to six billion of power assets, and roughly thirty billion of chips. Against that denominator, every second the site is not running at full output is an enormous amortized cost, so curtailing for maintenance or model changes is normal and curtailing because power is expensive is, in his words, never. That is what makes colocation a supply-side answer specifically. If the load will not bend, the generation has to come to it.