Starcloud CEO Philip Johnston says its first satellite, Starcloud One, cost about $2.5m to build against a quoted $75m to $100m.
Johnston argues space compute beats terrestrial solar data centers once launch costs fall below about $500 per kilogram, since space avoids land permitting and battery backup.
Starcloud has filed with the FCC for 88,000 satellites, and it is booked on Falcon 9 rideshares about 21 months ahead while planning Starship launches.
Starcloud bets cheap Starship launches will move data centers into orbit
Starcloud is betting that data centers will move to orbit. Once launch falls below about $500 per kilogram, solar power in space with no land, batteries or backup beats building on the ground. The company has shown it can build satellites cheaply and run GPUs in orbit, so its growth now depends on Starship delivering cheap, frequent launch capacity.
Takeaways
Orbit beats ground below $500 per kilogram. Johnston argues that once launch costs fall under about $500 per kilogram, space compute beats terrestrial solar data centers. Orbit needs no permitted land and no batteries or backup power, and each square metre of panel yields about eight times the energy.
Scrappy building cut satellite cost roughly fortyfold. Starcloud One cost about $2.5m against quotes of $75m to $100m. Adi and Ezra Feilden built it largely by hand and tested on a $200 secondhand vibration table. A new table costs over $1m with a 26-month lead time.
Launch access is the binding constraint. The company's first act was booking a Falcon 9 rideshare about 21 months ahead. Insurance cannot recover lost time, Feilden notes. Johnston expects Starship reusability within a year and commercial capacity from 2028, and Feilden praises Rocket Lab's Neutron, due in 2027.
GPUs in orbit already run paying workloads. Starcloud One flew the first NVIDIA H100s in space last November, alongside other vendors' chips. It trained Andrej Karpathy's nanoGPT and runs inference on radar and hyperspectral imagery for government customers, downlinking vessel coordinates in minutes instead of hours.
Cooling versus reliability splits the founders. Adi calls radiators the hardest problem, where cost and manufacturability matter more than feasibility, using the liquid cooling standard in data centers. Johnston says cooling is largely solved and reliability matters more, so Starcloud runs a three-month chip burn-in and duplicates pumps.
Scale target is 88,000 satellites. Starcloud has filed with the FCC for 88,000 satellites, each about two tons drawing 235 kilowatts. Fifty per Starship launch add about 10 megawatts. An auto-style assembly line should reach 100 satellites a week, with headcount growing from 30 to 100.
Starcloud Two in January is the proof. Starcloud Two has cleared vibration testing and spans about 120 feet, carrying the largest commercial deployable radiator in space. The company has raised $450m and earns about $1m a month in interest on its cash.
Johnston's predictions
SpaceX will buy Tesla once the stock unlocks.
Elon Musk's claim that SpaceX reaches $100tn in ten years is achievable if there is no world war.
Tap a name to go there · tap a line to see why they are linked
On this episode 7
Show 1
Optimus robots building Optimus factories will drive exponential growth in manufacturing.
Most orbital workloads will be inference for coding agents, back-office agents and chat, since large training clusters are unlikely to dock together soon.
Satellites moving at 17,000 miles an hour are hard to attack, and tampering with one would count as an act of war against the US.
The Fermi Explorer side project
A nonprofit cofounded by Johnston and others, including Camilo Lopez, plans to send a spacecraft to Alpha Centauri, about four light years away.
Estimated cost is $10m to $15m for a journey of roughly 80,000 years, using solar-powered Hall effect thrusters and only existing technology.
Payloads may include the golden record and a digital copy of a fruit fly brain.