Starcloud CEO Philip Johnston on launching AI chips into space

Summary

Philip Johnston, co-founder and CEO of StarCloud, discusses StarCloud 1, the first satellite to run and train AI models in orbit using an NVIDIA H100 GPU. He explains the case for space-based data centers: lower latency for spacecraft inference and, over time, abundant low-cost solar energy that could make orbital compute far cheaper than Earth-based power. Johnston also outlines a planned 5 GW orbital data center, addresses radiation and orbital-debris risks, and emphasizes a lean engineering-focused team.

• StarCloud launched its first spacecraft with an NVIDIA H100 and ran an AI model and LLM in orbit.
• Johnston says short-term use is inference and cloud compute for spacecraft to reduce downlink latency.
• Long-term thesis is moving most compute to space for abundant low-cost solar energy.
• He estimates all-in energy costs below 1 cent per kWh, about 10x cheaper than new North American projects.
• The company aims to build a 5 GW orbital data center, which would be hard to power on Earth.
• Radiation and orbital-debris risks are addressed through H100 accelerator testing, shielding, low orbit, and redundant thrusters.
• Johnston highlights a 12-engineer team and keeping the company lean and engineering-dense.