You guys need to hear this because it's one of those stories that keeps going in my head β€” SpaceX is planning a literal factory in orbit, not just rockets anymore. Musk revealed plans for a constellation of up to ONE MILLION satellites capable of generating 120 GigaWatts and powering potentially up to 100 million frontier-class GPUs while the world watches his public company's valuation climb on exactly that kind of news. The claim is that this isn't magic β€” it leverages existing Starlink V3 hardware, which Musk said in a recent video they don't think is a super hard problem at all. But Matt Desch from Iridium called out the hype directly during an earnings call, calling it a long-term opportunity with massive technical hurdles and questioning whether the push for orbital computing is genuinely needed or if someone just wants to tack another moonshot onto their valuation.

Let's talk about what this actually means in terms of hardware because the numbers are wild even before we get into orbit. Each AI1 satellite would pack solar panels covering roughly 600 square meters, which is about six basketball courts per unit and generates up to 150kW peak power while pulling an average of around 120kW for computing alone. Those panels probably weigh between one and two metric tons on their own β€” but that's the best part because you also need a radiator system large enough to dump the heat from the GPUs, which adds another at least one or two metric tons of hardware before we even count the bus, storage, or computers. So each satellite will land somewhere in the three-to-seven-ton range and get launched on an upper stage like Starship V3 with a 100-metric ton payload capacity to low Earth orbit β€” meaning SpaceX is already planning for multiple satellites per launch from day one.

Source: https://arstechnica.com/space/2026/07/how-hard-is-it-to-build-orbital-data-centers-actually/