Google and SpaceX in talks to put AI data centres into orbit, report says
TechCrunch reports, citing sources, that Google and SpaceX are in early-stage talks to build AI data centres in orbit. The conversations took place at the companies' Mountain View and Hawthorne offices and no specific financial agreement is yet in place.
The underlying problem is the hard cost math of building compute infrastructure in space. As of 2026 it costs roughly $2,700 per kilogram to lift hardware to low Earth orbit on a Falcon 9. Orbital data centres are still expensive against the $800-per-kilogram hardware cost of ground-based hyperscale sites.
Against that cost, orbit offers three concrete advantages. First, continuous solar power: solar panels in mid-to-high orbit generate energy 24 hours a day. Second, cooling: thermal radiation in a vacuum can be a thousand times more efficient than air cooling. Third, the growing environmental pressure on land-based data centres.
According to the Data Center Consortium's annual report, global data centre water use reached 1.2 billion cubic metres in 2025. Their carbon emissions are roughly a third of those of global aviation. These numbers are the central pressure behind Google's "carbon-neutral compute" strategy.
SpaceX's Starship-class rockets are crucial to any orbital data centre plan. Starship V3 is being designed for a 200-tonne payload. A single Starship launch could lift roughly 5 MW worth of data-centre hardware into orbit. By comparison, a small hyperscale site needs 50-100 MW.
Google's annual report indicates that the company operated roughly 8 GW of data-centre capacity for AI training and inference in 2025. That number is expected to reach 30 GW by 2030. If orbital solutions can carry even 1% to 2% of that capacity, the pressure on land-based sites is meaningfully reduced.
Professor Olivier de Weck, of MIT's space systems group, told the BBC: "Computing in orbit makes sense physically. The main remaining obstacle is service and repair costs. On the ground, swapping a GPU takes hours. In orbit, it needs either an astronaut or a robotic system."
China is moving in this direction too. China Aerospace Science and Technology Corporation (CASC) announced earlier this year a project called "Tiangong data-centre platform." The plan is to launch a prototype 5 MW orbital data-centre module in 2027. A US-China race in this area will pull investment in quickly.
On the financial side, Bank of America analyst John Hodulik shared the following figure in a client note: the orbital data centre market could become a $22 billion industry by 2032. That projection is small next to the roughly $350 billion annual ground-based market, but its growth rate makes it attractive.
On the regulatory side, there are no mature rules yet. The US Federal Communications Commission and the International Telecommunication Union are re-examining orbital spectrum and orbital placement rules. If the Google-SpaceX project moves forward, a 2028 target for the first pilot is plausible; commercial scale would be mid-2030 at the earliest.