Greetings, Earthlings: Philip Johnston of Starcloud on Data Centers in Space

Philip Johnston of Starcloud argues space will host the majority of new AI compute capacity within a decade, with all-in energy costs below half a cent per kilowatt hour.

• Space infrastructure costs under $5M/megawatt vs. $15–20M/megawatt terrestrial; all-in energy cost projected below half a cent per kilowatt hour at scale.
• Break-even launch cost for space vs. terrestrial data centers is ~$500/kg; rising permitted land costs push that threshold closer to $1,000/kg.
• StarCloud 1 has 5 Nvidia GPUs (including one H100) on orbit with zero chip restart failures; 70% of engineering time goes to heat dissipation, not radiation.
• Heat problem is solved via deployable liquid-cooled radiators scaling T^4—heat pumps boost 60°C chip fluid to 100°C radiator temp to maximize radiation output.
• StarCloud 3 design: 200kW per satellite, 50 units per Starship = ~10 MW per launch; first cost-competitive commercial payloads expected mid-to-late 2028.
• Current revenue comes from military edge inference (SAR satellites discard 90% of data today); military GPU time prices ~1,000x terrestrial rates.
• Johnston positions Starcloud as Equinix to SpaceX’s AWS: infrastructure-and-energy layer, customers supply their own chips and sell their own cloud services.
• Johnston predicts ~$1 trillion/year capex deployed in space compute within 10 years, calling it the largest market opportunity in history.

2026-03-17 · Watch on YouTube