ITU examines orbital AI data centres as operators file plans for tens of thousands of satellites

The International Telecommunication Union (ITU) examines the idea of orbital AI data centres in an article drawn from an episode of its Space Connect series. It says during 2026 operators have submitted filings covering tens of thousands of satellites, and for one proposal a million, which might expand computing and storage in space. Larger satellites and multi-satellite distributed computing concepts are also being developed.

Speakers describe differing motives. A researcher at the European Space Policy Institute calls Europe's interest a sovereignty play, while China wants to ease bandwidth limits and US firms want to extend their cloud lead. The European Space Agency's Ian Carnelli says security and defence are the current drivers, and a nearer opportunity is processing Earth-observation data onboard rather than training language models in space. NTT's Brandon Karpf says a single large platform faces heat and mass limits, while spreading compute across small satellites hits a link-speed gap, about 100 gigabits per second compared with around 7 terabits between chips on Earth. He tested five possible uses: physics ruled out two, economics ruled out two, and a sovereign cloud was the one left. He says bandwidth, not compute, is scarce, so business cases often hinge on spectrum decisions.

On regulation, DLA Piper's Julie Kearney says no jurisdiction has adopted a dedicated framework for orbital data centres. Operators must coordinate radio frequencies and orbits through the ITU, and questions also arise on cybersecurity, intellectual property, AI governance and data control. She says talks should continue within the ITU and at COPUOS, the UN's outer-space committee. Speakers also note that orbit would shift environmental pressure to launches, debris and re-entry, and Narda Jones says it does not solve the AI infrastructure problem.