The idea is simple: AI demands more energy than Earth’s grids can provide, so relocate data centers to space, where there’s perpetual sunlight and free electricity. SpaceX, Blue Origin, and numerous startups are competing to realize this concept. However, scientists and engineers point out the vision overlooks crucial aspects of thermodynamics, economics, and orbital mechanics.
SpaceX filed an application with the FCC on January 30 to launch up to one million satellites into low Earth orbit. Each satellite would include computing hardware, collectively offering “unprecedented computing capacity to power advanced artificial intelligence models.” These satellites would orbit 500 to 2,000 kilometers up, maximizing sunlight exposure, and utilize SpaceX’s Starlink network. SpaceX also requested a waiver for the standard deployment milestones, typically needing half a constellation operational within six years.
Blue Origin followed with its own filing seven weeks later. Project Sunrise suggests deploying 51,600 satellites in sun-synchronous orbits between 500 and 1,800 kilometers, aided by the previously announced TeraWave’s 5,408-satellite constellation for ultra-fast optical backhaul. Blue Origin highlighted its architectural focus, planning to perform computations in orbit and transfer results through TeraWave’s mesh network.
Startups move faster. Starcloud, rebranded from Lumen Orbit, raised $170 million with a valuation of $1.1 billion in March, achieving unicorn status in Y Combinator history only 17 months post-program. It launched a satellite with an Nvidia H100 GPU in November 2025 and applied with the FCC in February for a constellation of up to 88,000 satellites. Aethero, focusing on defense, develops space-grade computers with Nvidia Orin NX chips and raised $8.4 million, currently testing hardware in space.
The commercial logic addresses a real issue. Global data center energy usage reached 415 terawatt-hours in 2024, with projections exceeding 1,000 TWh by 2026 according to the International Energy Agency. AI server acceleration drives 30 percent annual growth. In Virginia, data centers use 26 percent of total electricity; Ireland might hit 32 percent by year-end. Grid constraints, permit delays, and resistance to expanding terrestrial capacity are significant.
However, orbital computing faces physical obstacles. The main issue is heat. In space, processors rely on radiative cooling, demanding large surface areas. Dissipating a megawatt of heat while maintaining electronics at 20°C requires about 1,200 square meters of radiators—around four tennis courts. A commercially viable data center needs radiators thousands of times larger than those on the International Space Station.
Radiation adds another challenge. Low Earth orbit exposes chips to cosmic rays, inducing bit flips and permanent damage. Radiation hardening increases costs by 30-50% and cuts performance by 20-30%. Triple modular redundancy, an alternative, involves trifold copies of each chip, cooling needs, electricity, and mass. Starcloud’s use of shielded commercial GPUs remains experimental, lacking proof of scale or longevity.
Latency is another issue. A million satellites at altitudes from 500 to 2,000 kilometers cannot achieve the required inter-node communication for frontier AI model training. In contrast, ground-based solutions achieve 10 to 50-millisecond latencies, suitable for inference rather than training, the major AI compute demand.
Cost is significant: IEEE Spectrum estimates a gigawatt data center in orbit would cost over $50 billion, thrice the cost of an equivalent terrestrial facility. Google claims launch costs must drop below $200/kg for space computing to become feasible. Current Starlink operations cost $1,000 to $2,000/kg. Analysts suggest a competition threshold of $20-30/kg, unattainable within the next 20 years.
The astronomical community raises additional concerns. SpaceX’s FCC application drew around 1,000 public comments opposing it. Many argued that it would outnumber visible stars, increasing space militarization and commercialization.
While orbital data centers may arise, SpaceX’s Starship could transform mass-to-orbit economics. Starcloud’s incremental approach of deploying small payloads and testing radiation resilience may lead to breakthroughs. Current terrestrial energy constraints persist.
Yet, bridging FCC filings for millions of satellites to competitive orbital computing involves complex physics challenges that current AI infrastructure investment cannot bypass. The issue isn’t the theoretical feasibility of space data centers but why they’re viewed as solutions to pressing, immediate problems when major engineering obstacles remain. The sky isn’t the limit—the radiator is.