Google has sent a Tensor Processing Unit into orbit aboard a prototype satellite built by Planet Labs, testing whether AI computing hardware can work in space. The trial puts real operating conditions, especially power and heat, ahead of the much bigger question of whether orbital data centres could ever make practical sense.
Watch Desk analysis
What happened
The satellite launched from California aboard a SpaceX rocket, according to The AI Insider. Google’s TPU is due to run AI models in 15-minute bursts, while the mission tests whether it can draw about a kilowatt of continuous power and stay cool in orbit.
Travis Beals, who manages Google’s Project Suncatcher, told the publication that ground tests cannot fully reproduce those conditions. Google’s next planned demonstration, described as a follow-up for next year, would link two purpose-built satellites by laser for heavier workloads.
The longer-term ambition is an orbital data centre of 81 satellites flying in close formation. Beals said the rockets needed to scale that idea cost-effectively do not yet exist. The report also says Google has released a peer-reviewed white paper, to be published in Joule, projecting that SpaceX could reach launch prices near $200 per kilogram by 2035. Its scenario would require about 1,800 Starship launches.
Why it matters
This is a test of the unglamorous requirements behind space-based computing: supplying power, managing heat and operating a chip reliably in orbit. Google’s reported plan to run inference workloads, rather than massive training runs, also puts a boundary around what this hardware is meant to demonstrate.
The 81-satellite vision is not the same thing as a buildable business plan. Launch costs, the scale of the required rocket programme and the results of this first trial all stand between a prototype and an orbital data centre. Space is a dramatic place to put a computer, but it does not grant free electricity or a convenient repair desk.
Our read
The first TPU in orbit is a concrete experiment, not proof that data centres are about to leave the ground. The useful result will be whether the chip can do sustained, valuable work under real orbital power and thermal constraints, and whether the costs can eventually come down enough to justify the spectacle.
What to watch
- Whether the TPU completes its planned tests and runs AI models reliably in orbit.
- What the planned two-satellite laser-link demonstration shows about heavier workloads.
- Whether launch costs approach the levels needed for Google’s larger vision.
Discussion spark: If orbital computing becomes technically viable, should companies pursue it despite the enormous launch and infrastructure demands, or should investment stay focused on improving data centres on Earth?
Sources and evidence
- Google Sends Its First TPU Into Orbit to Test Space-Based AI Compute (5 October 2026, 17:26 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.