🚀
Starcloud wants a 4-km, 5-GW GPU data center… in spaceNVIDIA-backed startup
Starcloud is about to put a
data-center-class GPU in orbit: their
Starcloud-1 satellite launches in November with an
NVIDIA H100 onboard — roughly
100× more GPU compute than any prior space system. The 60-kg, fridge-sized sat is a first step toward orbital GPU clouds.
Why it matters: today, satellites downlink raw data to Earth for processing. With an H100 upstairs,
AI runs where data is born — cutting response times from
hours to minutes. That’s big for sensors like
SAR, which can spit out ~
10 GB/s; instead of shipping it all home, the model can filter, segment, and alert
in orbit.
Power & cooling: sunlight is (nearly) constant in orbit and space is an
“infinite heat sink” via radiators, so no water-hungry chillers. Starcloud’s roadmap sketches a
~5 GW orbital data center fed by solar + radiator panels about
4 km × 4 km. Future birds aim to integrate
Blackwell GPUs for up to
10× more performance versus Hopper.
Biz timeline: Crusoe (the “energy-first” cloud) plans to deploy capacity on Starcloud’s platform from
2026–27. Starcloud projects
~10× lower energy cost than Earth data centers long-term. A white paper even models a
40 MW cluster with a
~$8.2M launch cost —
if Starship-class launch prices fall to ~$30/kg (a big if).
CEO Philip Johnston’s bet:
within 10 years, most new data centers get built in space. Gamers, rejoice — astronauts might finally run
DOOM at ultra.
#Starcloud #H100 #Blackwell #OrbitalCompute #SpaceDatacenter #SAR #EdgeAI
———
@science