Will data centers in space actually be cheaper than building them on Earth? According to Philip Johnston, co-founder and CEO of Star Cloud, the answer is yes — and the break-even point is closer than most people think. Johnston recently presented the economic and engineering case for orbital computing, arguing that once launch costs drop to around $500 per kilogram, space-based data centers will undercut terrestrial ones on pure cost. For context, SpaceX's Starship is designed to hit $10–$20 per kilogram. We're not talking science fiction. We're talking about a decade, maybe less.
0:45
Star Cloud 1 separation confirmed — the first H100 GPU successfully deployed to orbit
Watch at 0:45 →
Star Cloud has already put hardware in orbit. Their first satellite, Star Cloud 1, carried five NVIDIA GPUs including the H100 — the most powerful AI chip currently available — and used it to train a model in space for the first time in history. The company has now filed with the SEC for a constellation of 88,000 satellites, each producing around 200 kilowatts of power, collectively delivering roughly 20 gigawatts of compute capacity. Johnston calls it "just scratching the surface."
Will Data Centers in Space Be Cheaper Than on Earth?
The economic argument starts with a simple comparison to the cheapest form of energy we have on Earth: solar. If you want to build a solar-powered data center on Earth today, you face three major costs:
- Permitted land — In North America, this is often the single largest cost for a new solar project.
- Battery storage and backup power — Because peak solar output lasts only about four hours per day, you need massive battery banks to run through the night.
- The solar panels themselves — The actual hardware, which is surprisingly the smallest of the three costs.
Now compare that to building the same project in space. You pay zero for land. You pay zero for battery storage because a satellite in the right orbit is in sunlight 24 hours a day, 7 days a week. And you need eight times fewer solar panels because one square meter of solar panel in space produces eight times the energy of one square meter on Earth's surface.
The only major additional cost in space is launch cost. That's it. And as launch costs continue to fall — driven largely by SpaceX's Starship program — a clear break-even point emerges. Johnston pegs that crossover at around $500 per kilogram of payload. Starship is designed to eventually deliver $10–$20 per kilogram. The math is going to flip, and it's going to flip hard.
3:20
Concept render of the 88,000-satellite constellation in dawn-dusk sun-synchronous orbit
Watch at 3:20 →
How Do Space-Based Solar Panels Power a Data Center?
The constellation Star Cloud is currently building operates in what's called a dawn-dusk sun-synchronous orbit. This orbital path keeps the satellite permanently on the terminator line — the boundary between day and night on Earth's surface — which means it's always facing the sun. No orbital night. No power interruptions. No batteries needed.
Each satellite in the planned 88,000-unit constellation generates around 200 kilowatts of power. Across the full constellation, that adds up to roughly 20 gigawatts of compute capacity. The satellites are connected via optical laser links and can serve inference workloads to anywhere on Earth with sub-50 millisecond latency — fast enough for most real-world AI applications including code generation, business process agents, and media synthesis.
The total capital expenditure for this constellation is estimated at around $100 billion — which Johnston notes is actually lower than what it would cost to build equivalent terrestrial infrastructure. That's the point at which this stops being a curiosity and starts being a genuine infrastructure play.
Can GPUs Actually Run in Space?
Until recently, many experts thought you simply couldn't run consumer-grade, state-of-the-art AI chips in space. Two problems seemed insurmountable: thermal dissipation and radiation tolerance. Star Cloud 1 disproved both.
The satellite carried NVIDIA's H100 GPU and successfully used it to train a model (specifically, Andrej Karpathy's NanoGPT), run inference on synthetic aperture radar (SAR) satellite imagery, and execute a version of Google's Gemini model — all from orbit. Jensen Huang highlighted the collaboration at NVIDIA's GTC event, where he also revealed that NVIDIA and Star Cloud are co-developing a new chip called Rubin 1, specifically designed for space operation. The key innovation: it runs at higher temperatures without increased failure rates, which directly reduces the mass required for thermal radiators.
How Do You Keep a Space Data Center From Overheating?
This is one of the trickiest engineering problems in orbital computing. Space is cold — about 3 Kelvin — but it's also a vacuum. That means you can't use convection or conduction to dump heat. The only option is thermal radiation in the infrared spectrum.
Everything warm emits infrared radiation — including humans. The challenge is scaling that up to handle hundreds of kilowatts of waste heat from a GPU cluster. Johnston laid out the math clearly: a solar panel in space produces roughly 200 watts per square meter. A radiator panel running at 50°C can dissipate around 800 watts per square meter. That means for every 400 square meters of solar panel, you need about 100 square meters of radiator — a 4:1 ratio.
The key lever here is the Stefan-Boltzmann equation, which states that thermal radiation scales with the fourth power of temperature. A modest 10% increase in the radiator's operating temperature in Kelvin can nearly halve the required radiator surface area. That's why the hotter-running Rubin 1 chip matters so much: higher chip temperature equals smaller radiator equals less mass equals lower launch cost.
9:45
Concept render of a 5-gigawatt, 4km x 4km training structure assembled in orbit from Starship launches
Watch at 9:45 →
Is Kessler Syndrome a Real Threat to Space Data Centers?
Kessler syndrome — the theoretical chain reaction of satellite collisions creating a debris field that makes orbit unusable — is a legitimate concern that Johnston takes seriously. But he argues it's less of an immediate threat than it appears, for a few reasons.
First, Star Cloud's initial satellites fly at around 400 km altitude. At that height, the atmosphere is thick enough that any debris naturally de-orbits within months. A collision wouldn't create a permanent debris field — it would clean itself up. Second, higher orbits where debris could persist longer are actually very sparsely populated, because they edge into the Van Allen radiation belts, which are hostile to electronics. Third, the visual representation of orbital congestion is deeply misleading: on those crowded-looking satellite maps, each dot is approximately the width of California representing an object the size of a refrigerator. Space is vast.
Johnston points to SpaceX operating roughly 10,000 Starlink satellites without a single collision as a real-world proof point. Sophisticated collision avoidance software, not orbital scarcity, is the actual solution. And he estimates that the dawn-dusk sun-synchronous orbit alone could accommodate terawatts of compute capacity without meaningful collision risk.
Does Radiation Cause Bit Flips That Break Chips in Space?
Yes, radiation is a genuine concern — and Star Cloud addresses it through exhaustive ground testing rather than exotic space-rated hardware. The team has run four rounds of testing at a cyclotron in Knoxville, Tennessee, a high-velocity proton particle accelerator that simulates the radiation environment of low Earth orbit. For heavy-ion radiation, they use the Brookhaven National Laboratory's facilities.
In a single 24-hour test session, they can expose chips to the equivalent of five years of radiation dose in orbit. The resulting telemetry data informs both shielding design and software-level decisions — like how aggressively to use error-correcting code memory, how to schedule redundant computations, and which chips are worth flying at all. It's a methodical, data-driven approach rather than hoping that consumer silicon survives.
What Is a Dawn-Dusk Sun-Synchronous Orbit and Why Does It Matter?
A sun-synchronous orbit is a polar orbit timed so that the satellite always crosses the equator at the same local solar time. The dawn-dusk variant keeps the satellite on the terminator line between day and night, which has two massive advantages: the solar panels always face the sun, providing continuous power, and the satellite avoids the Earth's shadow entirely. No eclipse. No power interruption. No batteries.
For a data center, this is transformative. Continuous power means continuous compute. Combined with optical inter-satellite links, the constellation can route workloads globally with sub-50 millisecond latency — competitive with many terrestrial cloud regions — while drawing on effectively unlimited solar energy. Johnston describes this architecture as the foundation not just of the largest infrastructure project in human history, but of what he calls a Kardashev Type II civilization: one that harnesses the full energy output of its star. Whether or not that framing resonates, the near-term business case is compelling enough on its own.
