Orbital AI data centers unlikely before 2030, Forethought finds
Original title: Will we put data centers in space?
Cost parity needs launch near $250/kg versus about $1,500/kg today, and only if Starship reuse matches Falcon.
Several companies have announced plans for AI compute in orbit, including Google’s Project Suncatcher, a SpaceX FCC filing for up to one million data-center satellites, and NVIDIA’s Space-1 Vera Rubin Module. Forethought’s model says space solar reaches parity with present-day off-grid power at roughly $250/kg to orbit, from about $1,500/kg today, and undercuts terrestrial energy near $50/kg. Radiative cooling looks manageable, but failed chips cannot be swapped and bandwidth likely limits early use to inference. The authors estimate orbital centers are unlikely to be a meaningful share of compute before 2030 unless Starship stays on track.
Unconfirmed: Cost and timeline estimates assume Starship stays on track and no transformative AI.
- Google announced Project Suncatcher for TPU satellites in dawn-dusk sun-synchronous orbit.
- SpaceX filed with the FCC for up to one million data-center satellites.
- China’s Three-Body constellation has 12 satellites and has run Qwen3 in orbit.
- About 38% extra non-compute hardware may be needed over five years.
- NVIDIA announced the Space-1 Vera Rubin Module at GTC in March 2026.
- Organizations
- FCCxAIBlue Origin
- Countries
- China
- AI models
- Qwen3
From the original article
14 min read at forethought.org
Several major technology companies have announced plans to operate AI data centers in orbit. Elon Musk recently claimed : “the lowest-cost place to put AI will be space […] within two years, maybe three.”
By simonebrunozzi · forethought.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.