Inside The Startup Launching AI Data Centers Into Space

Starcloud co-founders Philip Johnston, Ezra Feilden, and Adi Oltean explain how they launched the first Nvidia H100 GPU into orbit in 15 months and why orbital data centers could undercut terrestrial compute costs.

• Starcloud launched the first H100 GPU into orbit on Nov 2, 2025 — 100x more powerful than any prior space computer.
• Their break-even launch cost model requires ~$500/kg; space-based solar needed $50/kg and was abandoned after the 95% energy transmission loss made it unviable.
• Orbital data centers use zero fresh water: cooling is infrared radiation into deep space, not evaporation — which is draining rivers near US terrestrial data centers.
• Target product is a 40 MW, ~100-ton orbital data center fitting one Starship payload bay, running on uninterrupted sun-synchronous solar.
• Second satellite launches October 2026 with Nvidia Blackwell architecture, 10x more compute than StarCloud-1, and 24/7 optical terminal connectivity.
• Core IP is a large, low-mass deployable radiator — half the engineering team builds it; co-founder Ezra Feilden holds a PhD and worked on NASA lunar pathfinder deployable panels.
• Google, SpaceX, and Amazon are now independently exploring orbital data centers, validating the market direction Starcloud entered first.

2025-11-13 · Watch on YouTube