StarCloud wants to solve AI's energy and water crisis by launching data centers into orbit, where solar power is constant and heat dissipates into deep space.
The Earthbound Bottleneck
AI's appetite for compute is colliding with the planet's finite resources. Data centers consume vast amounts of electricity and fresh water—evaporating millions of gallons daily to cool servers—while straining grids already struggling with electrification. In parts of the United States, data centers are literally sucking rivers and lakes dry. Yet demand is only accelerating. The western world, meanwhile, has lost its ability to build large infrastructure projects quickly. Permitting, land acquisition, and grid interconnection can take a decade or more.
StarCloud co-founder Philip Johnston saw this tidal wave coming. Rather than fight for resources on Earth, he asked a different question: what if you moved the data centers where energy is abundant and cooling is free? Into space.
From Space-Based Solar to Orbital Compute
The original idea was space-based solar power: giant panels in orbit beaming energy down to Earth. But the numbers didn't work. Transmitting power through the atmosphere loses about 95% of the energy, meaning the launch cost would need to drop to roughly $50 per kilogram to break even—a target still decades away. So Johnston and his co-founders flipped the logic. Instead of sending power down, why not send the compute up?
Re-running the math gave a much more plausible threshold: $500 per kilogram. That's within striking distance today, thanks to SpaceX and startups like Stoke Space that are reusing rockets daily and scaling production. Launch capacity could increase by a factor of 100 or 1,000 this decade, making orbital data centers economically viable far sooner than space-based solar ever could.
The Hard Technical Problem: Heat
Critics have been vocal. The most common objection is thermal management: in the vacuum of space, there is no air to carry heat away. Dense GPU clusters generate enormous heat, and dissipating it requires large surface area—which seems impractical. Johnston acknowledges the challenge but points to his co-founder Ezra, a PhD engineer who spent a decade building large deployable structures for NASA's Lunar Pathfinder mission. Half of StarCloud's engineering team is dedicated to a single piece of IP: a low-cost, low-mass deployable radiator.
Instead of evaporating water, the system uses infrared radiation into deep space as its heat sink. A fluid circulates through the radiator panels, but never leaves the loop. The radiator simply sheds heat into the cold void. It's an elegant solution that bypasses the water scarcity issue entirely—a massive advantage over terrestrial data centers, especially in drought-prone regions.
A Team Built for the Bet
Johnston's own background is in software and applied math, not aerospace. But he compensated by assembling a co-founding team that covers every critical gap. Addi, a principal software engineer from SpaceX with 20 years of data-center experience at Microsoft, handles the compute module and chip radiation hardening. Ezra, the satellite-structures expert, builds the spacecraft bus and deployables. Between them, the three cover commercial compute payload, satellite structure, and the business case.
That complement allowed them to move with unusual speed. From founding to orbit took 15 months—about four times faster than the typical space startup. The first satellite, StarCloud 1, is a 60-kilogram spacecraft about the size of a small fridge, carrying an Nvidia H100 GPU, custom antennas, and the deployable radiator. Onboard, it will run the first AI training in space, fine-tune models, and even run a version of Google's Gemini. It's 100 times more powerful than any computer previously operated in orbit.
Why the Bet Could Pay Off
Johnston is open about the risk. 'It's easier to build a hard company than an easy one,' he says, paraphrasing Sam Altman. The logic is counterintuitive: if there's one enormous technical hurdle, and you clear it, everything else becomes easier—hiring, fundraising, press coverage. The market risk is minimal because demand for AI compute is insatiable. The only question is whether StarCloud can operate data centers in space cheaply enough.
Their second satellite, planned for October 2025, will be at least ten times more powerful, using Nvidia's Blackwell architecture and laser-based optical terminals for continuous 24/7 connectivity. The long-term vision is a 40-megawatt orbital data center—about 100 tons, fitting in a single Starship payload bay. If that works, the implications are enormous: zero fresh water usage, zero grid strain, and near-limitless expansion because the only bottleneck is launch capacity, not land or energy.
The Birth of an Industry
StarCloud is no longer alone. Google, SpaceX, and Amazon are all exploring orbital data centers. The November 2024 launch—StarCloud 1's separation confirmed—may be remembered as the moment a new industry began. Johnston and his team are betting that within a decade, most new data centers will be built in space, not on Earth. It sounds wacky to many. But Y Combinator didn't flinch when they pitched the grand vision, and now they have a satellite overhead proving the concept one inference at a time.
'Anything that's worth doing is going to be hard,' Johnston says. 'If something is too easy, it probably doesn't have the same potential outcome.' That willingness to chase a small-probability, massive-impact outcome is precisely what makes StarCloud's bet worth watching—and worth taking.