In Today’s Issue: What the ground is actually short of, what SpaceX says and what it spends, why Google's case is a long-dated bet, what Europe, the Gulf and China want from orbit, and which work will fly first. The record as of October 2.

Dear Readers,

"Orbital compute" covers two quite different ideas. One is to put computers on satellites so they can analyse their own camera and radar images in orbit and send down results instead of raw data. The other is far more ambitious: data centres in space, powered by sunlight, that run AI for customers on Earth. Google's Project Suncatcher and SpaceX's plans for its xAI unit belong to the second kind. This issue is about that second kind, with the first as a comparison: it shows where computing in orbit already has an obvious advantage.

Its case is easy to state. Data centres need enormous amounts of electricity, and in the right orbit a solar panel sees the sun almost around the clock. In the week around October 1 the idea moved: Starship reached orbit for the first time, the US Federal Communications Commission (FCC) took satellite operations out of federal environmental review, and Google's first Suncatcher test satellite reached orbit (SpaceWatch Global, 09/2026; Via Satellite, 09/30/2026; Google, 10/01/2026).

Chips in orbit are still expensive, hard to cool and hard to repair. So it is fair to ask what the companies expect to gain, and whether it is really electricity:

Why are Google, SpaceX and their rivals putting AI chips into orbit, which problem on the ground are they trying to escape, and what will those chips do up there first?

All the best,

Kim Isenberg

None of the three events settles whether computing in orbit can pay. Starship's first orbital flight ended hours early, according to accounts of the webcast, and there was no booster catch and no reuse, the steps that would make launches cheap (ScienceTimes, 09/29/2026; SpaceX, 09/2026). Google's satellite, built with the operator Planet, is a single test of whether its chips survive launch, radiation and heat; a test of laser links between two satellites, promised for early 2027, now has no month attached (Google Research, 11/04/2025; Google, 10/01/2026). And the FCC change is smaller than it sounds, because satellite licences had been exempt as a class from routine environmental review since 1986 (Payload, 09/2026).

In August, in "The Orbit Charges Twice," we looked at what orbital data centres would cost. This issue asks the step before: what each company says it is solving, set against what it actually does with its money.