AWS Doubts Economics of Space Data Centres
Google and SpaceX are pursuing orbital AI computing, but AWS's chief executive expects cost and cooling barriers to persist for the next decade or two.
Updated on Nov. 15, 2026 in Data Centers

Amazon Web Services CEO Matt Garman said space data centres are highly unlikely to be economically feasible or practical in the next decade or two. The idea remains at the research and proposal stage, not a deployed product.
Why it matters
The debate reflects a broader push to find power and cooling for AI hardware, as Google and SpaceX pursue plans to run AI chips in orbit. Garman's objections centre on launch cost and frequency, equipment weight, heat removal, and maintaining servers in space.
The evidence is an executive assessment paired with project plans and a filing, not measured orbital performance. AWS deploys one GPU every 14 seconds in terrestrial facilities, while Google's Project Suncatcher plans two prototype satellites for early 2027 and SpaceX seeks authorization for up to one million satellites. Whether the economics close at scale has not been demonstrated.
The players
Amazon Web Services
The cloud computing arm of Amazon, operating large terrestrial data centre networks and deploying AI accelerators at scale.
Alphabet's technology company, developing its custom Tensor Processing Units and running the Project Suncatcher orbital computing research.
SpaceX
Elon Musk's launch and satellite company, which has applied to the FCC for an orbital data centre system of up to one million satellites.
Matt Garman
Chief executive of Amazon Web Services, who questioned the near-term economic feasibility of space data centres.
Sundar Pichai
Chief executive of Google, associated with the company's Project Suncatcher orbital computing effort.
The details
Orbital data centres would place processors on solar-powered satellites, with Google's Project Suncatcher exploring networks carrying its Tensor Processing Units. Future satellite clusters would exchange data through laser links, allowing their processors to work together on larger AI workloads. The main engineering hurdles are cooling, since space has no air to carry heat away and dissipating it is quite hard, and the cost and weight of launching equipment.
Timeline
November 2025: Google announced Project Suncatcher.
January 2026: SpaceX applied to the US Federal Communications Commission for permission to launch and operate up to one million satellites as an orbital data centre system.
September 2026: Google described an initial hardware test in a project update.
Early 2027: Google's planned launch of two prototype satellites with Planet.
The Tech Race
Google's Project Suncatcher and SpaceX's FCC application put two competing orbital computing efforts on the table, with Google testing heat pipes, radiators and chip radiation resistance. AWS's position is that terrestrial sites such as deserts, with access to solar energy and cooling, remain the more practical choice for now.
Orbital data centres are not available to users, and no timeline for commercial service has been announced. For now, the practical effect is on how cloud providers plan and site AI computing capacity on the ground.
The takeaway
Watch whether Google's two prototype satellites launch in early 2027 and report results on heat removal and chip radiation resistance. Those findings will show whether the cooling and cost objections Garman raised are being overcome.
What happens next
Google's two prototype satellites are scheduled to launch in early 2027, the first live test of the Project Suncatcher concept.
Further reading
For more on the infrastructure behind this debate, see our Data Centers coverage.


