The hardest part of a computer to get is no longer the computer. It is the electricity.
This week Google puts its own AI chips on a SpaceX rideshare to see whether they survive the trip. The aggregator posts call it "AI data centers in space are actually happening." Google's own sentence says the mission is "designed to gather in-orbit data on how our TPUs handle the physical stress of spaceflight and the radiation and thermal extremes of space."
That is a stress test with a launch date, not a data centre.
The pitch is the sun: up to eight times the power of the same panel on Earth. That part is real. So is the reason anyone is looking. The constraint on AI stopped being chips and became megawatts, and the queue for power on Earth is now long enough to make a radiator in a vacuum worth testing. The number that makes it competitive is a launch price under 200 dollars a kilogram. Google's own paper says maybe the mid-2030s.
The Nvidia digital-twin work I developed for the BMW and Mercedes EV factories had one purpose. Find the failures on the cheap copy before you pay for the real one. Google has the same rule in a different medium. Fly one before you fly a thousand, and publish the test list.
Why it's Material. For the next decade this is a money story about electricity on Earth. The number to watch is your power bill, not the sky.
The Monday move. Ask whoever owns your AI spend one question. What happens to our price when the vendor's power bill goes up.
Monday Morning Mashup No. 139. What moved in AI, every Monday.
Video: Google, official, 24 September 2026.
