The renderings
The thing you have to understand about the renderings is the light. Not the satellites — the light, the way the orbital data centers are always shown at that specific terminator-line hour where everything is backlit gold and the radiator fins throw these long clean shadows and the solar arrays are doing the thing arrays do in the renderings, which is to say catching a sun that in the image has no glare, no thermal load, no problem¹ — and you look at it for a while, the way you'd look at a car ad, knowing it's a car ad, and you still want the car. That's the first tell. The wanting arrives before the analysis, and the analysis, when it shows up, is mostly just the wanting wearing a lab coat.
So here is the claim, and then here is why the claim is already a little dishonest. The claim: orbital compute is a folly — a structure built to be looked at, not occupied — and the actual money is at ground level where nobody's pointing the camera.² I believe this. But notice what the claim does for the person making it, which is to position him as the sober one, the adult in the room who isn't fooled by the gold light, and there is a specific pleasure in being the unfooled person that is itself a kind of foolishness, a vanity that I am, right now, by naming it, attempting to launder into credibility — which is also a move, and you should probably distrust this whole paragraph, including the part where I tell you to distrust it.
What the renderings leave out, in no particular order: the 100 kW of waste heat that has to go somewhere and in vacuum can only radiate, slowly, off surfaces that have to be the size of a tennis court; the launch cost that is quoted as a target the way a diet is quoted as a target; the fact that the satellite can talk to other satellites at roughly a terabit per second but can only get maybe one to ten gigabits down through the actual weather to an actual ground station that an actual person has to maintain in, like, a field somewhere with a fence around it; the GPU that will be obsolete in two years bolted to a chassis designed to last six; the regulators; the ITU; the spectrum; the slow grinding committee-shaped reality that the gold light has no pixels for.³ The models that price all this produce numbers with two decimal places. The decimals are the tell. Nobody who actually knew would dare be that precise.
And but so here's the part I keep circling, the thing under the thing: I think the orbital story is so seductive precisely because it lets us not look at the ground, where the real build-out is happening, which is unglamorous and enormous and is mostly about power and heat and zoning and the patience of balance sheets, and which photographs about as well as a substation. The crowd looks up because up is where the beauty is. The load gets transferred down here, quietly, at the seam, while everyone's neck is craned.
I don't actually know when the folly becomes a building. Maybe 2040. Maybe never, because every year the ground gets a little better — cheaper power, better cooling, storage falling toward fifty dollars a kilowatt-hour — and every increment down here pushes the parity line further out there, which means the engineers working hardest to make orbit happen are, by improving the alternative, the very people quietly killing it. There's something almost tender in that, if you let yourself feel it: a whole industry straining toward the sky, and the thing that defeats them is their own competence at staying home.
¹ A rendering is a promise rendered as a fact, which is the same grammatical move as a forecast, which is the same move as a prospectus, and you start to notice that an enormous amount of capital allocation is just people responding to verb tense.
² "Nobody" is wrong, obviously — SemiAnalysis is pointing the camera there, several serious people are, I'm aware of the irony of claiming contrarian sight in a note that cites the consensus skeptics. Fine. Read "nobody" as "not the marginal dollar," which is the only nobody that moves a price.
³ This is the catalog the style demands and I'll be honest that there's a pleasure in the accumulation itself, in piling the unglamorous specifics until the sheer mass of them outweighs the gold light — which is a rhetorical trick, the list as argument, and it works on me even though I'm the one doing it.