communityfix.org

Cape Canaveral, Florida, USA

#00222

PartialGlobal

Case study of

#00213 Relocate compute into orbit, powered by continuous sunlight and cooled by radiators

Implementer

Starcloud (formerly Lumen Orbit), satellite bus by Astro Digital, launch by SpaceX

Timeline

Since Nov 2, 2025

Location

Cape Canaveral, Florida, USA28.4900, -80.5800

Description

Starcloud-1 launched on a Falcon 9 from Cape Canaveral on 2 November 2025 into a 325 km orbit: a 60 kg satellite on an Astro Digital Corvus-Micro bus carrying a single Nvidia H100, with a design life of approximately 11 months followed by controlled deorbit. The satellite ran inference workloads and a nanoGPT-scale training run (company-sourced, unverified independently). What was demonstrated: a single accelerator can be launched, powered, thermally managed, and operated in LEO for months. What was not demonstrated and is not addressable at 60 kg: multi-kilowatt radiative heat rejection, inter-satellite optical fabric, multi-year radiation survival, servicing, or economics at any scale. Nvidia's CEO described orbital data centre economics as 'poor today' on the 26 February 2026 earnings call.

Metrics

5
Satellite mass60kg
Accelerators flown1Nvidia H100
Orbit altitude325km
Design life~11months
Capital raised to date200,000,000USD

Funding

Venture capital; $200 million raised in total, including a $170 million Series A led by Benchmark and EQT at a $1.1 billion valuation

Lessons learned

  • A single-unit orbital demonstration validates survivability, not scalability. The constraint that decides the proposition — radiative heat rejection — does not bind at 700 W and dominates at megawatt scale; any replication attempt must address this before scaling.
  • Distinguish the demonstration from the roadmap: the flown hardware and the stated 5 GW ambition differ by roughly seven orders of magnitude in power, making the gap a core due-diligence item.
  • The most defensible near-term orbital compute application is processing data that already originates in orbit (avoiding downlink of raw sensor data), which is a different proposition from relocating terrestrial AI capacity and should be scoped and funded separately.

Documented Jul 28, 2026

Author AvatarGerard Antoun

communityfix.org