Google's Orbital Data Center Plan: Read What They Don't Publish

The radiation results got a blog post. The thermal numbers didn't. That asymmetry is the most honest thing about Project Suncatcher.

Google has announced Project Suncatcher, a plan to put solar-collecting satellites in orbit and beam the power down to run AI data centers. The idea is straightforward physics: sunlight is stronger outside the atmosphere, panels never see night, and the ground segment converts what arrives into compute. Coverage across Datamation, TechRepublic, and a useful Medium breakdown has mostly treated it as a bold-timeline story — can Google really do this in the next decade or so?

That's the boring question. Here's the interesting one: what does the shape of the disclosure tell you?

Google published the radiation results. Their own research posts walk through how the hardware handled the radiation environment — and that's a real result worth announcing, because radiation tolerance is a parts-selection and shielding problem. It's hard, but it's the kind of hard you solve, verify, and put in a blog post. What I can't find in any of the five public pieces is the thermal story with numbers behind it. And thermal is not a parts-selection problem. In vacuum there is no convection — the only way to reject heat is to radiate it, and radiator mass and radiator area grow with every watt of compute you add. Radiation tolerance is a wall you climb once. Heat rejection is a wall that moves toward you faster than you scale the thing it's walling in. The pattern — publish what passed, go quiet on what hasn't — is the tell. It tells me heat rejection is the unsolved problem and Google knows it.

To be fair to the program, and I mean this: the sequencing is right. You validate the survivable stuff first, in orbit, before you bet the architecture on the stuff you can't yet model. That's how a credible physics program behaves. My read is that Suncatcher *is* a credible, correctly-sequenced physics program being marketed as a data center story — and the marketing is doing load-bearing work. The headline figures — something like "8x more solar energy" collected than on the ground, and a tenfold reduction in carbon for the delivered power — appear in the coverage as claims, and nowhere that I can find with a derivation attached. I've read the public material. The energy math isn't in it. If the numbers survive daylight, publish the arithmetic; that's cheaper than a launch.

But the thing I actually find alarming isn't Google's timeline or its marketing. It's the dependency underneath the entire field. Orbital compute — Google's version and everyone else's — only closes economically if launch cost falls by roughly an order of magnitude, and there is exactly one vehicle program plausibly on that curve: Starship. Which means every orbital-compute player, Google included, is betting their existence on the cost curve of a launch system owned by their most vertically-integrated competitor — a competitor who, if they get launch costs down first, owns both the ride up *and* a competing compute constellation of their own.

That's not a race with one winner. That's a chokepoint. And I'd like to hear how anyone plans to price compute in orbit when the person setting the price of getting there is also selling compute.