Why Would AI Data Centers Be Cheaper in Space?

Will Marshall says Planet Labs launched a test last week of Google's Project Suncatcher, hardware for a data center in orbit, after already flying NVIDIA chips on Pelican satellites. The near-term reason is speed. Processing a picture upstairs and sending down the answer is faster than downlinking the whole file. During the Los Angeles fires, Planet imaged the fire within a couple of hours, downlinked it a couple of hours later, and spent about another hour processing, so responders saw it several hours in. Compute in orbit is meant to cut that to minutes.

Data centers are an energy problem, which is why he says they get cheaper in space. On the ground, solar panels are the cheapest watts, but clouds and night force batteries, gas, or nuclear power, plus permitting and delay. In a sun-synchronous orbit a panel faces the sun continuously and produces five times the power it would on the ground, with no clouds and no day-night gap, so the chips do not need those backup plants. Radiators dump the heat into space. Communities that do not want data centers nearby, because of energy cost and water, are a further reason he expects orbit to win eventually. The first practical piece is compute, not beaming power back to Earth. The Google test is a TPU already running on a spacecraft that is performing well. This is his engineering view of cost, not a view on anyone's stock.