Space Based Data Centers are a Scam.

“Put all the GPUs in space” sounds clever until you run the numbers. Let's estimate the cost of launching and sustaining "all the GPUs" in orbit.

Space isn’t a freezer. In vacuum you can’t convect heat away because you can only radiate it, which means enormous, fragile radiator surfaces for MW-scale clusters. That radiator area becomes a micrometeoroid/debris risk and a control/engineering headache.

Power is worse. A serious AI cluster is multi-MW. In orbit that means huge solar arrays + storage (eclipses) or nuclear (politically/operationally hard), plus you still have to dump the waste heat.

Then there’s the business reality: GPUs refresh every ~2–3 years. On Earth you swap parts daily; in space, upgrades and repairs are “space mission” problems. Add radiation-induced errors (cosmic ray bit flips), bandwidth/latency limits vs fiber, launch mass costs, and debris risk, and the economics flip negative very obviously and very fast.

Space compute can make sense only for niche on-orbit processing (e.g., satellites that must process sensor data before downlink). For general data-center AI: Earth wins on power, cooling, maintenance, and cost.

Jon Peddie Research projects an installed base reaching ~3,008 million GPUs. Even if you pretend each GPU averages only 0.2–0.5 kg and is designed for space applications and ignore everything else, *launch-only* is:

3.0B × (0.2–0.5 kg) × $2,720/kg ≈ $1.6T–$4.1T launch-only

At a mid value (≈250 W/m²), 8 GW of compute heat needs about 32 km² of radiators. Using 5 kg/m² implies ~160,000 tons of radiators alone.

…and that’s before the manufacturing cost, failed launches, power, structure, comms, R&D, and the fact you can’t practically operate billions of separate “space PCs" (cost scale factor increase on the most *extremely unrealistically* conservative end of *at least* 10x or $160T).

So when you do these basic back of the envelope calculations, it becomes obvious that the "climate change" "it's more efficient in space" "do it to conserve the water" "for the future of humanity" arguments are pure marketing garbage designed to conceal the true purpose for these projects which is probably that they just want to conceal data from *you* or need to collect some data that only exists in space, and their project is more important to them than building things you actually need.

Operationally it’s also absurd: there are on the order of ~14,300 active satellites today. “3 billion space PCs” would be ~200,000× more spacecraft than exist, with impossible spectrum/ground-station, tracking, and debris/collision constraints. (For perspective, Starlink is ~9,400 satellites and already dominates the active satellite population.)

And with all those rocket launches... you have the "environmental impact" of that as well. Why fund things people actually need on Earth when you have all the monopoly money?

In either case, they are just lying to you and insulting your intelligence.