Interview, Fireside Chat
The Case For Data Centers In Space
- StarCloud, founded January 1, 2024, by co-founder and CEO Philip Johnston, raised a $170 million Series A led by Benchmark, becoming the fastest-growing unicorn in YC history 17 months after Demo Day.
- The company builds data centers in space to solve terrestrial energy bottlenecks for AI, leveraging abundant solar energy and rapidly declining launch costs (projected to reach $500/kg breakeven).
- StarCloud pivoted from "Space-based Solar" (Lumen Orbit) in late 2023 to orbital data centers after calculating that transmitting solar power to Earth incurred 95% energy loss, whereas running compute in space had a viable $500/kg launch cost breakeven.
- The team booked their first SpaceX rideshare launch on January 2, 2024, 18 months in advance, establishing a "forcing function" to align hardware development with launch dates before finalizing payload specifications.
- StarCloud 1 launched in November 2025, carrying five GPUs (two NVIDIA, two Arm, and one NVIDIA H100) submerged in phase-change material for thermal management, a novel approach validated after a 24-hour ice-bath thermal cycling test conducted by the founders.
- Following launch, the team resolved critical software issues requiring manual ground station intervention over three days, achieving first contact in 12 hours and completing the two-week commissioning period to run the first Gemini model in orbit.
- StarCloud 2 (10kW) is scheduled for 2026 to provide commercial compute for government and military customers via 3 optical terminals for high-bandwidth downlink of raw satellite imagery.
- StarCloud 3 (planned for ~2028, 200kW, 3 tons) will deploy 50 units per Starship to generate 10MW of compute per launch, targeting hyperscale data centers.
- The company filed an FCC application for a constellation of 88,000 satellites in dawn-dusk orbit, theoretically capable of generating 20 gigawatts of initial capacity and scaling toward 10 terawatts (20x the US power grid).
- Engineering challenges are addressed by a team of 20 engineers (half former SpaceX/NASA), focusing on two core problems: thermal dissipation in a vacuum (using radiators 500x cheaper than ISS standards) and radiation tolerance.
- To mitigate radiation bit-flips, StarCloud avoids expensive space-grade components, instead rigorously testing automotive-grade off-the-shelf electronics in particle accelerators (Brookhaven National Lab, Knoxville cyclotron) to validate shielding and software mitigation.
- StarCloud is developing a custom "Rubin Space" chip in partnership with NVIDIA, involving stripping non-essential components from H100s and stiffening boards for space deployment.
- StarCloud 2 will also fly a Bitcoin mining ASIC and AWS Outpost hardware to support military customers requiring edge inference and local instances of cloud compute.
- The company initially faced rejection from over 100 VCs and YC itself, citing the "sci-fi" nature of the idea and skepticism regarding low-cost launch viability and radiation feasibility.
- Investor sentiment shifted following public market corrections in software stocks, rising global energy constraints, and regulatory bans on new terrestrial data centers (e.g., in New York), making space-based compute a strategic national security asset.
- StarCloud partners with SpaceX for Starlink laser terminals on the next 20 satellites to solve interconnectivity, addressing high-bandwidth data transmission requirements.
- Future roadmap includes deploying to polar orbits (starting StarCloud 4) to achieve 24-hour sun exposure, eliminating the need for orbital drag compensation thrusters required for mid-inclination maintenance.
- Philip Johnston, originally a software engineer and McKinsey consultant, recruited co-founders Addy (ex-SpaceX) and Ezra (ex-NASA) based on a hypothesis that launch costs would drop 10x, prioritizing world-class engineering talent over initial business models.