Space Data Centers: Can Solar Energy in Orbit Really Work?

SpaceX promises that Low Earth Orbit will provide a near-limitless source of solar energy for its satellite data centers dedicated to artificial intelligence computing. Named Starmind or AI1, Elon Musk’s project aims to place up to 1 million satellites at an altitude of 600 km. Each unit would generate 120 kW using 600 m² of solar panels, exploiting the continuous solar radiation available in orbit. But infrared radiation cooling, cosmic radiation, and the limited lifespan of components turn this energy dream into a technical nightmare. In a document filed with the Securities and Exchange Commission (SEC) in June 2026, SpaceX acknowledges that these infrastructures rest on “unproven technologies” that “could never become commercially viable.”

Space Energy: An Illusory Abundance for Data Centers?

The central argument of the Starmind project rests on constant access to solar radiation. At an altitude of 600 km, satellites escape Earth’s day-night cycles and enjoy an energy flux of 1,367 W/m² with no atmospheric attenuation. With 600 m² of solar panels, each unit could theoretically generate 120 kW continuously. A terrestrial data center consumes on average 30 to 50 MW for 10,000 servers, about 3 to 5 kW per rack. On paper, a Starmind satellite would thus power about twenty racks’ worth of equipment. But the energy equation ignores a critical parameter: the heat removal from the processors.

Deploying 600 m² of solar panels in orbit represents a formidable mechanical challenge. By comparison, the International Space Station (ISS) carries 2,500 m² of panels for 420 kW, but its total mass reaches 420 metric tons. Current Starlink satellites weigh about 260 kg and measure 3 meters long. A Starmind satellite, equipped with giant panels, high-performance processors and massive cooling systems, could weigh several tons. SpaceX’s Starship rocket can place 100 tons into Low Earth Orbit, but launch cost and deployment complexity rise with the mass. The energy architecture of terrestrial data centers benefits from decades of optimization, which is absent in space.

The Orbital Data Center Cooling Nightmare

On Earth, data centers use water, air or free cooling to dissipate heat. In orbit, the vacuum of space forbids convection or heat conduction. Only infrared radiation can evacuate thermal energy. Yet a modern processor emits between 200 and 400 W per chip. To cool thousands of chips effectively, satellites would need giant radiators capable of emitting heat as electromagnetic radiation. The efficiency of this process depends on the radiator surface area and their temperature. The larger the surface, the heavier the mass. A space data center thus becomes a perpetual compromise between compute power and cooling capacity.

The radiators needed to evacuate heat from high-performance AI processors could amount to several hundred square meters per satellite. NASA estimates that a space radiator dissipating 1 kW requires about 1 m² of surface. For a satellite generating 120 kW and producing equivalent heat, radiators would reach 100 to 150 m², in addition to the 600 m² of solar panels. The total mass climbs, launch costs explode. According to Ars Technica’s calculations, deploying 1 million satellites would require about 3,500 launches per year for five years, i.e., nearly 10 launches per day. At $10 million per Starship launch, total costs would range from $1.45 trillion to $10 trillion depending on scenarios.

Cosmic Radiation: The Invisible Enemy of Electronics

Low Earth orbit satellites are constantly bombarded by cosmic radiation and solar particles. Processors, memories and electronic circuits degrade progressively. Energetic particles cause bit flips, logic malfunctions and accelerated transistor wear. Starlink satellites burn up in the atmosphere in under five years(2), releasing a cocktail of pollutants. Unlike terrestrial data centers where failed servers are replaced within hours, no maintenance is possible in orbit. Each satellite becomes obsolete after a maximum of five years, forcing a perpetual renewal of the fleet.

Modern AI processors, like Nvidia H100 GPUs or Google’s TPUs, contain billions of transistors etched at 4 or 5 nanometers. This extreme miniaturization makes them particularly vulnerable to radiation. Space-grade components use hardened technologies (radiation-hardened), but they lag several generations technologically and cost ten to a hundred times more. A terrestrial GPU in 2026 outperforms a radiation-hardened space processor from 2020. To keep AI calculations competitive, SpaceX would have to renew its entire constellation every five years, i.e., 200,000 satellites per year. This planned obsolescence driven by physical constraints makes the business model untenable.

Beyond energy and technical challenges, the Starmind project raises major environmental questions. Mass launches and the disintegration of satellites threaten the ozone layer(3) and inject pollutants into the stratosphere. A coalition of organizations, including DarkSky International, PEER and Earthjustice, filed a petition with the U.S. Federal Communications Commission (FCC) requesting a moratorium on new mega-constellation licenses. Low Earth Orbit risks becoming an environmental sacrifice zone, sacrificed for the sake of an energy promise whose viability remains unproven. Can space solar energy truly power AI, or is it a technological mirage masking an impending ecological catastrophe?