The orbital data center pitch runs like this: launch costs are finally low enough to make it work. Run the numbers on Starship payloads, stack up the satellite cost models, and the spreadsheet starts to close. That's the story everyone is telling. It is missing the hard part.
Power availability in orbit is not a line item in the launch-cost spreadsheet. It is the actual problem.
Here is the physics. A data center module in low Earth orbit swings between direct solar flux and shadow roughly every 90 minutes. When it is in the sun, it is absorbing roughly 1,361 watts per square meter — 36 percent more than reaches Earth's surface, since the atmosphere is not in the way. When it is in shadow, it is dissipating heat into a vacuum that conducts and convects essentially nothing. Radiators are the only cooling path. And according to SatNews, dissipating just one megawatt of heat while keeping electronics at a stable 20°C requires a radiator surface of approximately 1,200 square meters — roughly the size of four tennis courts. A 100-kilowatt rack — a unit that would fit in a small room on the ground — needs 120 square meters of radiator. The International Space Station, all 420 tonnes of it, produces less than 100 kilowatts average from its solar arrays. Scale that to a data center that matters and the thermal management problem is not an engineering footnote. It is the dominant design constraint.
The power generation side looks better, at least on paper. A solar panel in orbit produces roughly five times the electricity the same panel would produce on Earth, according to SpaceNews. No atmosphere, no night, no weather. Google Research's Project Suncatcher notes that in the right orbit a solar panel can be up to eight times more productive than on the ground. That math is real. The question is whether the cells can survive long enough in the LEO radiation environment to matter.
The current generation of space-qualified solar panels — the silicon-based arrays on the ISS, which are about 14 percent efficient and produce roughly 190 watts per square meter — are not the answer. Perovskite cells are the more interesting prospect. According to published research, perovskite solar cells achieve specific power of 23 to 30 watts per gram, a 10- to 15-times improvement over conventional silicon arrays at 0.5 to 2 watts per gram. Current flight-ready prototypes have demonstrated up to 1,960 watts per kilogram. That is a real advance. But perovskite stability in the LEO radiation environment — the combination of galactic cosmic rays, solar particle events, and total ionizing dose over years — remains unproven at scale. The technology is early-stage, and space-qualification timelines are measured in years, not quarters. The SpaceNews piece treats perovskite as a near-term enabler. That framing deserves skepticism.
The nuclear path has more near-term credibility. Zeno Power, a startup developing radioisotope power systems, raised a $50 million Series B in May 2025 and holds more than $60 million in contracts with the U.S. Department of Defense and NASA, according to SpaceNews. The company's approach uses strontium-90 fuel sourced from the Department of Energy — a fission product with a 28-year half-life — to generate electricity via a Stirling engine. No chain reaction, no moving parts except the piston, no cooling system requiring radiators the size of tennis courts. Zeno says it will launch a demo satellite in 2026 and expects its first commercial nuclear battery by 2027. That timeline is plausible. But the fuel supply chain, safety certification path, and regulatory approvals for new radioisotope systems are not trivial. NASA has used radioisotope power for decades, but those systems used plutonium-238 — a different fuel, a different supply chain, a different approval process.
There is also a conflict worth naming plainly. The SpaceNews opinion piece that sparked this discussion is by Oleg Demidov, a general partner at Beyond Earth Ventures. Beyond Earth Ventures is an investor in both Zeno Power and Tandem PV, which the piece mentions approvingly. This is not hidden — Demidov discloses it in his author bio. But it means the article's optimistic framing on perovskite and nuclear power is also an investment thesis from someone with direct financial interests in two of the companies named. Readers should know that.
The terrestrial power situation adds another layer. The International Energy Agency estimates global data center power consumption could double to 945 terawatt-hours annually by 2030, from 415 TWh in 2024, roughly equivalent to Japan's total electricity consumption, according to the IEA. A Department of Energy report found data centers consumed about 4.4 percent of total U.S. electricity in 2023, with projections reaching 6.7 to 12 percent by 2028. Turbines are sold out through 2030. Power purchase agreements for new nuclear capacity are being signed in blocks of hundreds of megawatts, per SpaceNews. On the ground, power is the bottleneck too. The idea that orbital data centers sidestep this constraint assumes the power problem is solved at the satellite level. It is not — it is different and, in some ways, harder.
The cost math from IEEE Spectrum gives the clearest picture of where the economics actually stand. A one-gigawatt orbital data center — roughly the scale of a medium-sized terrestrial facility — would require approximately 4,300 satellites, each with a 1,024-square-meter solar array generating 250 kilowatts. The total cost including launch and five years of operations: roughly $51 billion. A comparable terrestrial system over the same period: about $16 billion. The orbital premium is real, and at three times the cost it is better than the 7-to-10-times estimates from earlier models. But "better than before" and "economically competitive" are different statements. The $51 billion figure also depends heavily on Starship reaching its target launch economics. Falcon 9 prices today are roughly $3,600 per kilogram to low Earth orbit. Starship's fully reusable configuration is targeting $100 to $500 per kilogram. The range is wide. The gap between current pricing and target pricing is where the business case either works or does not.
China is not waiting for the economics to sort themselves out. ADA Space launched 12 Star Compute craft into low Earth orbit in March 2026 as the first nodes of a planned 2,800-satellite Three-Body Computing Constellation, according to Per Aspera. Each satellite carries an 8-billion-parameter model with 744 teraoperations per second of compute and 100 gigabits per second optical cross-links. This is a different kind of bet — state-backed, not venture-funded — and it changes the risk calculus for Western competitors. The question is not just whether orbital data centers make economic sense. It is whether the strategic value of having compute infrastructure outside another country's jurisdiction justifies costs that would kill a purely commercial project.
Jensen Huang, Nvidia's chief executive, acknowledged the thermal challenge directly at a March 2026 conference. "In space, there is no convection, there is just radiation, and we have to figure out how to cool these systems out in space," he said, according to CNBC. That is not a dismissive quote. It is an engineer acknowledging a real problem. Nvidia's Vera Rubin Space-1 Module, designed specifically for orbital AI workloads, is real hardware — but it is being designed into systems that do not yet exist at commercial scale.
The orbital data center story is not about whether the technology will eventually work. Some version of it probably will. The story is about the gap between what the launch-cost narrative implies and what the power infrastructure actually requires. Solar arrays that survive long enough. Radiators that fit within mass budgets. Nuclear systems that clear regulatory review. Fuel supplies that do not create new bottlenecks. None of these are unsolvable. All of them take time, money, and engineering that the current hype cycle is not pricing in. The people running the launch-cost spreadsheet are not wrong. They are just solving the wrong problem.
Notebook: The supply chain for space power — fuel isotopes, radiation-qualified cells, power electronics — is itself a nascent industry with its own bottlenecks. Watch who controls the supply of strontium-90 from DOE inventories, who has the testing infrastructure for LEO radiation qualification, and whether the solar cell manufacturers can produce at the specific powers needed without sacrificing operational lifetime. That beat is worth its own story.
