Rogue Time Lord One

Quick note for the reddit physicists who jumped in to "correct" @broudsov: they answered a different question. Yes, you can cool electronics in space. Satellites have done it for sixty years, at watts and kilowatts. The real question here is whether you can cool and power a gigawatt of GPUs easily and cheaper than a building in Texas can. Agreeing with the first says nothing about the second. The second is what matters because that's what @elonmusk is ultimately selling. Here's the physics. On Earth, chips dump heat into air or water, and contact carries it away. Space has nothing to touch. Heat can only leave by radiation, one photon at a time, obeying Q = εσAT⁴. At chip-safe temperatures an ideal radiator gives you a few hundred watts per square meter. The ISS, the most sophisticated thermal system ever flown, rejects up to 70 kW with 422 square meters of ammonia radiator. Call it 166 watts per square meter, flight-proven. A single NVIDIA GB200 rack draws 120 kW. Sit with those two numbers for a second. And the scaling is linear. Not approximately linear. Linear. Every watt of compute needs the same fixed patch of radiator, forever. The only lever in the equation is temperature to the fourth power, and silicon caps your temperature around 85°C. There's no clever move left. More compute means more square meters, more mass, more launches. "Scaling is not a problem" is exactly backwards. Scaling is the entire problem, and it never gets better with size. "Starship makes launches cheap" doesn't beat the equation. It just changes who pays. The EU's own orbital data center study (Thales, Airbus, HPE) budgets about 2.5 square meters of radiator and roughly 5 square meters of solar panel per kilowatt. A 1 GW facility is 2.5 square kilometers of radiator and 5 square kilometers of array, tens of thousands of tonnes before you count coolant, pumps, and kilometers of plumbing, all riding the same rocket. The startup white papers admit it too: 4 km structures, hundreds of launches just to build the thing. Then maintenance shows up. No, GPUs don't all die at year three. Plenty run longer. But at gigawatt scale, some fraction is failing all the time, and here's what happens on Earth when one does: a human walks down the aisle, finds the card, pulls it, seats a new one. In orbit, that human doesn't exist. You launch a machine that can locate one dead card at a grid coordinate inside a rack, extract it, and seat the replacement, in vacuum. That machine rides the rocket too. And when the servicing robot breaks? You launch a robot to fix the robot, or you write off the rack. If the answer is Optimus, fine. Now Optimus lives in a radiation environment and somebody has to service Optimus. It's maintenance all the way down. "But the power is free." No. The sunlight is free. The collector isn't. Panels ride the rocket too, at 40-100 watts per kilogram of flight hardware, so a gigawatt of orbital solar is 10,000 to 25,000 tonnes of array. Then it starts dying on you. The best space-grade cells lose 10-15% of their output per decade in LEO from radiation alone. Cheap silicon loses 35-40%. Add atomic oxygen chewing surfaces, micrometeorites, 16 thermal cycles a day. NASA just spent $103 million bolting eight new arrays onto the ISS because the original wings aged out, and that was to keep 215 kW alive. A gigawatt needs 4,600 times that power system, with panel replacement launches on a schedule, forever. And none of it works until the silicon survives LEO. Trapped protons and cosmic rays flip bits and kill chips. Commercial GPUs aren't rad-hard, so you add shielding: tonnes per module, which barely stops the heavy ions anyway, forces ECC and redundancy overhead that eats your effective compute, and sits like a winter coat between the chip and the radiator you were already short on. Every problem on this list makes the others worse. Now notice the pattern. Cooling? Launch radiators. Power? Launch panels. Dead GPU? Launch a robot. Robot breaks? Launch another robot. Every single answer is "put it on a rocket," because the man selling the answer owns the rocket. When all you have is a hammer, every problem looks like a nail. When all you have is Starship, every data center looks like it belongs in orbit. Add it up the way a CFO would. Radiator mass, array mass, shielding mass, coolant, pumps, robots, replacement GPUs, replacement panels, launch after launch. Every line item lands in the price per compute-hour. On Earth, air and water carry your heat for free, a technician with a screwdriver does your maintenance, and if you want dedicated clean baseload you build nuclear. Thorium, if you like. No rocket required. Even paying full price for power, terrestrial compute wins by a mile. The orbital pitch only survives while nobody prices it. References: Reddit rebuttal to @broudsov reddit.com/r/elonmusk/commen… Lumen Orbit Whitepaper (obvious VC money grab) lumenorbit.github.io/wp.pdf EU ASCEND Study cordis.europa.eu/project/id/…