Johnston says space could provide abundant solar energy as building new power projects on Earth becomes more constrained. He began exploring space businesses after seeing the launch capacity planned for Starship, then shifted from space-based solar to data centers: he says transmitting power to Earth would lose about 95% of it, while his team estimates space data centers could break even at launch costs near $500 per kilogram.
The Case For Data Centers In Space
Speaker
Philip Johnston, co-founder and CEO of Starcloud, explains the company’s plan for space-based data centers on Y Combinator’s The Light Cone with the show’s host.
For its first spacecraft, Starcloud flew five GPUs, including an NVIDIA H100. Johnston describes using phase-change material to manage heat and ground testing to study radiation effects; the team also worked through software restarts after first contact. He says the spacecraft trained a model, ran Gemini, and performed inference on satellite imagery.
Johnston identifies heat removal in a vacuum and chip reliability under radiation as the main engineering challenges. Starcloud is developing deployable radiators and testing commercial components, while a SpaceX contract is intended to provide laser connectivity for its next 20 satellites. He describes a planned 10-kilowatt spacecraft for government and military customers, followed by a 200-kilowatt design, with the larger constellation and its capacity dependent on launch costs and Starship.
Johnston says investors initially rejected the idea, in part because they doubted launch costs would fall and found the concept difficult to picture. He credits his co-founders’ space engineering experience and says the team booked a rideshare launch the day after founding to create a concrete deadline. Early customers include government and military organizations, which could use onboard processing to turn large volumes of satellite imagery into information such as vessel locations.
Takeaways
- For a space startup, an early launch booking can turn an open-ended idea into a schedule and a minimum viable mission.
- Choose the first product around the constraints it avoids: Johnston says data centers do not require returning products to Earth.
- Test the hardest technical risks directly, including radiation effects on off-the-shelf components and heat management in vacuum.
- Sequence products around customers and capability: Starcloud plans to serve government and military users before pursuing hyperscale compute.
- Build the founding team around the technical work the company must solve, especially when the founder lacks that engineering background.
Mentioned in this talk
Search inside this talk
What was claimed 16
- usesrideshare launch
“we booked the first available SpaceX rideshare launch”
YouTube Unverified1 source· Talk
- built
“Starcloud is building data centers in space”
YouTube Unverified1 source· Talk
- metricStarcloud: Raised $170 million in a round led by Benchmark
“they raised $170 million, led by Benchmark”
YouTube Unverified1 source· Talk
- ceo ofPhilip Johnston CEO of StarcloudCEO
“the co-founder and CEO of Starcloud”
YouTube Unverified1 source· Talk
- foundedPhilip Johnston founded Starcloud
“Philip Johnston, the co-founder and CEO of Starcloud”
YouTube Verified2 sources· Talk
- launchedStarcloud launched Starcloud-1
“you launched Starcloud-1 in November 2025”
YouTube Verified2 sources· Talk
- featureStarcloud-1: Carried an NVIDIA H100 GPU
“The most interesting one was the NVIDIA H100”
YouTube Unverified1 source· Talk
- usestwo Arm GPUs flew on Starcloud-1
“five GPUs on there: three from NVIDIA and two from Arm”
YouTube Unverified1 source· Talk






