Training Data Greetings, Earthlings: Philip Johnston of Starcloud on Data Centers in Space
123 snips
Mar 17, 2026 
Philip Johnston, founder and CEO of StarCloud, builds space-based data centers and led the first StarCloud satellite. He explains why falling launch costs make orbital compute cheaper, how radiative cooling and solar power work in vacuum, and the thermal and networking designs that scale to gigawatts. He also previews latency tradeoffs, security and regulatory angles, and the larger economic shift toward space-based compute.
AI Snips
Chapters
Transcript
Episode notes
Pezz Dispenser Units Enable Rapid Scaling
- Modular small satellites (200 kW each) let Starcloud pack many units per Starship, yielding ~10 MW per launch and rapid capacity growth as Starship scales.
- Philip expects 50 units per Starship, 50×200 kW = ~10 MW, enabling gigawatts per month once Starship flies at rate.
LEO Inference Latency Matches Earth Needs
- Space inference latency matches Starlink-level performance, making most inference workloads feasible from LEO.
- Philip says sub-50 ms round-trip latency is achievable, so voice agents, chat, and video generation can run from space nodes.
Marginal Costs Flip In Favor Of Space
- Marginal cost dynamics invert: terrestrial marginal cost rises with each new data center while space marginal cost falls with manufacturing and launch scale.
- Philip predicts a crossover where new buildouts shift to space, estimating near $1T/year CapEx to space within a decade.