Data Centers in Spaaaaace - Space Tech Firms Want to Take Nvidia H100 GPUs Into Orbit to Power the Next Generation of Compute

Why It Matters

If viable, space‑based GPUs could dramatically lower AI compute costs and decouple large‑scale training from Earth’s energy constraints, reshaping the cloud‑computing landscape; yet the project's success hinges on overcoming formidable engineering and financial challenges.

Summary

Redmond‑based Starcloud, backed by Nvidia Inception, and cloud‑computing firm Crusoe are partnering to build the first orbital data centers, launching Nvidia H100 GPUs aboard a Starcloud satellite in November 2025 with limited AI compute services expected by early 2027. The design relies on unfiltered solar power in space, which the companies claim could cut energy costs up to tenfold compared with terrestrial facilities. By using the vacuum of space as an “infinite heat sink,” the venture aims to deliver gigawatt‑scale, renewable‑powered AI workloads without atmospheric losses. However, launch expenses, hardware maintenance, radiation exposure, and the lack of convection cooling present significant technical and economic hurdles.