Google’s Orbital Bet: Why Space Data Centers Are Still a Thermodynamics Nightmare

Posted on by

(SeaPRwire) –  

The pitch is seductive but physically illiterate. Google claims it is taking the data center race off-planet to escape terrestrial resource constraints. They launched Project Suncatcher in November 2025 to test this theory. It is a clever PR move, but the physics remains stubborn. Gravity is not a regulatory hurdle you can lobby away. The real enemy is heat, and space offers no breeze.

Officially, Google says the satellite operates as expected on October 1. It carries four TPU chips and uses solar panels. The mission is to test radiation and thermal extremes in low Earth orbit. They will keep the chip running for 15 minutes at a time. Then they shut it down to cool off. That is the headline number. Fifteen minutes of work followed by idle time. Compare that to a terrestrial server farm running at 100% utilization for days. The efficiency gap is massive.

Industry subtext suggests this is a test of viability, not deployment. SpaceX filed FCC plans for up to 1 million data-center satellites in early 2026. Blue Origin filed for 51,600 satellites in March. Musk said the lowest-cost place for AI will be space within two to three years. Google’s senior VP James Manyika was more realistic. He told the New York Times this will take years. He compared it to Google’s 15-year research period before AI breakthroughs. The market is buying the hype. The engineers are writing the checks.

Starcloud launched an Nvidia H100 chip last November. It ran Google’s Gemma model. Cowboy Space Corp. launched Reason-1 on the same rocket as Google. It tests laser power transmission to Earth. These are all signal tests. None are scalable compute. The supply chain landscape is clear. You can move silicon to orbit. You cannot move energy density to orbit without massive, expensive solar arrays. The launch cost is down to $1,000 per kilogram. But the heat rejection hardware is heavy. You are paying to move weight into a vacuum that offers no cooling. The orbital data center is a lab experiment, not a business model. It will remain a niche for specialized, low-duty-cycle workloads until thermal management solves the radiator problem. Do not expect a general-purpose compute layer in LEO this decade. The physics does not care about your funding.