I keep coming back to one number and it wont leave me alone, ten gigawatts, thats what a five month old startup called Orbital just told the FCC it wants to put in low earth orbit, on June 24th, barely two weeks ago, a filing for up to 100,000 satellites stacked with GPUs and radiators the size of small buildings, and the company has raised exactly five million dollars in pre seed money, which in this industry is basically lunch money, and I sat there reading the filing thinking, this is either the start of something enormous or its a very expensive way to learn about thermal engineering the hard way.

Silicon, stacked and stupidly expensive, singing softly under a sun that never sets on it, thats the pitch.

No clouds.

No permitting board.

No neighbor filing a complaint about the hum.

Here is the thing nobody wants to say plain, terrestrial power is the wall, the grid queues stretch past 2030 in some regions, communities are pushing back, cranes stalled, capital circling, cables clogged with permits nobody signed, and so the money has decided, apparently, that space is easier than a zoning meeting, which if you think about it for more then a second is a genuinely strange sentence to type out loud, and I will admit right here that when I first read the filing I got a little too EXCITED about a company that has flown exactly zero satellites, and I had to close the laptop for a minute and go stare at a wall.

Cowboy Space raised 275 million in May. Starcloud hit unicorn status off 170 million.

Now tiny, unproven Orbital wants a slice of the same sky, and Elon Musk already filed for one million satellites of his own, calling it, I am not making this up, a step toward becoming a Kardashev civilization.

The satellites will rise, thin silver birds against the skies, and somebody somewhere is already counting the money before the metal even flys.

And look, I got to be straight with you, so much of this doesnt work yet, not even close, Starcloud only has one H100 chip up their and even that struggled, the radiator wasnt strong enough to let it ran full power, according to IEEE Spectrum's recent teardown of the whole sector, which points out that hauling a million satellites to orbit on Starship, sixty at a time, needs something like sixteen thousand launches, and building them at ten times Starlinks pace would eat up twenty five years, longer then any of these companies put in a slide deck.

So heres me, sitting with the filing open in one tab and the Spectrum piece open in another, and I genuinely do not know which one is lying to me, or if theyre both just telling different parts of the same expensive truth, the one where a five million dollar startup and a trillion dollar rocket company are both betting, in their own separate ways and at wildly different scales, that the cheapest place to run an AI model five years from now wont be a warehouse in Virginia but a box in the sky, and I keep circling back to that thought, turning it over, unsure if its visionary or if its just the same old story tech tells itself about escaping the ground it stands on.

Not great, but here we are.

I think I need to lie down.