The allure of moving data centers into orbit is compelling: Artificial Intelligence (AI) demands more energy than terrestrial grids can provide, and space offers near-constant sunlight and free electricity. Companies like SpaceX, led by Elon Musk, and Blue Origin, spearheaded by Jeff Bezos, are racing to realize this dream. Nonetheless, experts warn that the ambitious vision overlooks crucial chapters in thermodynamics, economics, and orbital mechanics that remain unwritten.
On January 30, SpaceX submitted a proposal to the Federal Communications Commission (FCC) to launch as many as one million satellites into low Earth orbit, each equipped with computing hardware intended to create a constellation with "unprecedented computing capacity to power advanced artificial intelligence models." These satellites would be positioned at altitudes between 500 and 2,000 kilometers, designed to maximize sunlight exposure and utilize the existing Starlink network for routing traffic. SpaceX also requested a waiver for the FCC's standard deployment milestones, which typically mandate that half of a constellation must be operational within six years.
Just seven weeks later, Blue Origin applied with its Project Sunrise, which envisions deploying 51,600 satellites in sun-synchronous orbits ranging from 500 to 1,800 kilometers. This project is complemented by TeraWave, a previously announced constellation of 5,408 satellites that will provide ultra-high-speed optical backhaul. While SpaceX's proposal focuses on sheer scale, Blue Origin emphasizes a more sophisticated architecture that facilitates computation in orbit and transmits results back to Earth through TeraWave's mesh network.
The startup landscape is evolving rapidly as well. Starcloud, previously known as Lumen Orbit, raised $170 million at a valuation of $1.1 billion in March, becoming the fastest unicorn in Y Combinator history just 17 months after its inception. In November 2025, the company launched its first satellite featuring an Nvidia H100 GPU and filed with the FCC in February for a constellation of up to 88,000 satellites. Meanwhile, Aethero, a defense-oriented startup, is developing space-grade computers using Nvidia Orin NX chips wrapped in radiation shielding, having secured $8.4 million for its venture and currently testing hardware in orbit.
The commercial rationale for these ventures is grounded in real challenges. Global electricity consumption by data centers reached approximately 415 terawatt-hours in 2024, with the International Energy Agency projecting it could surpass 1,000 TWh by 2026, driven largely by AI servers that are expected to grow at an annual rate of 30%. In Virginia, data centers already account for 26% of the total electricity supply, while in Ireland, this figure could reach 32% by the end of the year. These statistics highlight genuine constraints in grid capacity, permitting delays, and political resistance to expanding terrestrial infrastructure.
However, scientists caution that the physics of orbital computing poses significant hurdles at any scale. The primary challenge is heat dissipation. In space, the absence of air means that heat generated by processors is removed solely through radiative cooling, which necessitates vast surface areas. To maintain electronics at a stable 20 degrees Celsius while dissipating just one megawatt of thermal energy, approximately 1,200 square meters of radiator space is required—equivalent to four tennis courts. For a commercially relevant data center operating at several hundred megawatts, the radiators needed would be thousands of times larger than those currently deployed on the International Space Station.
Radiation is another major obstacle. Satellites in low Earth orbit expose unprotected chips to cosmic rays and trapped particles, leading to potential bit flips and permanent circuit damage. Hardening hardware against radiation can increase costs by 30 to 50% and reduce performance by 20 to 30%. Alternatively, employing triple modular redundancy necessitates launching three copies of every chip, which triples the cooling requirements, electricity consumption, and mass. Starcloud's strategy of using commercial GPUs with external shielding is an intriguing experiment, but no one has yet demonstrated its viability at scale over extended hardware lifetimes.
Latency also poses a significant constraint. A constellation of a million satellites spread across altitudes of 500 to 2,000 kilometers cannot achieve the tight coupling necessary for frontier model training, where inter-node communication latencies must remain in the microsecond range. Low Earth orbit introduces minimum latencies of several milliseconds for inter-satellite links and 60 to 190 milliseconds for round trips from ground to orbit, compared to just 10 to 50 milliseconds for terrestrial content delivery networks. This makes orbital infrastructure potentially suitable for inference workloads, but not for training, which constitutes the majority of current AI compute demand.
Cost remains a critical factor as well. According to estimates from IEEE Spectrum, a one-gigawatt orbital data center could exceed $50 billion, approximately three times the cost of a comparable terrestrial facility, including five years of operation. Google has indicated that launch costs must drop below $200 per kilogram for space-based computing to become economically viable. Currently, SpaceX's Starlink operations are priced between $1,000 and $2,000 per kilogram, with analysts suggesting that the true competitive threshold for terrestrial alternatives lies around $20 to $30 per kilogram—a target that is not realistically achievable within the next two decades.
Even Sam Altman of OpenAI, who considered a multibillion-dollar investment in rocket manufacturer Stoke Space as a competitor in the orbital data center space, has publicly deemed the concept "ridiculous" for this decade. He has pointed out that the economics of launch costs versus terrestrial power costs simply do not align currently and raised the practical issue of how to repair a broken GPU in space.
Furthermore, the astronomical community expresses its own concerns. A substantial number of the nearly 1,000 public comments on SpaceX's FCC application urged the commission not to proceed, warning that the proposed constellation would surpass the number of visible stars in the sky for extended periods, further militarizing and commercializing an already congested orbital environment.
Despite these challenges, it does not entail that orbital data centers will never materialize. SpaceX's Starship, if it meets its cost targets, could revolutionize the mass-to-orbit economics that currently render the concept impractical. Additionally, Starcloud's incremental strategy of launching small payloads and improving radiation performance represents a potential engineering pathway to breakthroughs. The constraints posed by terrestrial grids driving interest in these projects are not going away.
However, the gap between submitting an FCC application for a million satellites and achieving economically competitive orbital computation compared to a facility full of GPUs in Iowa is not measured in mere years. It is a gap defined by complex physics problems that the current pace of AI infrastructure investment cannot bypass, regardless of how many billionaires are eager to pursue it. Scientists are questioning not whether space data centers are theoretically possible, but why, given the vast unresolved engineering challenges, anyone considers them a viable near-term solution to a problem that demands immediate answers. Ultimately, the sky is not the limit; the radiator is.