We wrześniu 2024 roku inżynierowie z firmy Starcloud argumentowali, że orbitalne centra danych są wykonalne i niezbędne dla rozwoju sztucznej inteligencji. W styczniu 2026 roku SpaceX Elona Muska podzieliło się planami wystrzelenia miliona satelitów w tym celu. Na Ziemi centra danych zużywają ogromne ilości energii i wody, zajmują rozległe tereny i budzą protesty lokalnych społeczności. Przykładowo, w maju 2026 roku rada gminy w Michigan wprowadziła moratorium na dostarczanie wody do hiperskalowych serwerowni. Kosmos wydaje się idealnym rozwiązaniem – energia słoneczna, naturalne chłodzenie i brak sąsiadów. Problem w tym, że chłodzenie w próżni jest nieefektywne, panele słoneczne wymagają ogromnych powierzchni, a wyniesienie miliona satelitów na orbitę kosztowałoby setki miliardów dolarów. Naukowcy są sceptyczni – w pełni funkcjonujące orbitalne centrum danych nie powstanie w tej dekadzie.
In September 2024, engineers from Starcloud in Redmond, Washington, published a white paper arguing that orbital data centers are feasible and necessary for AI development. In January 2026, Elon Musk’s SpaceX shared plans to launch one million satellites for that purpose.
Data centers on Earth are generating increasing controversy. The vast server farms powering artificial intelligence consume huge amounts of electricity and water, and take up large tracts of land. Local communities and politicians are taking action against them. For example, a township board in the US state of Michigan voted in May 2026 (a week before the article was published) to impose a one‑year moratorium on providing water to hyperscale data centers, so that it could study the impacts of a proposed facility.
A concept known since 2024
The discussion about using orbital data centers to power artificial intelligence is not new. In September 2024, engineers from Starcloud, a space technology company based in Redmond, Washington, published a white paper arguing that orbital data centers are feasible, economically viable and necessary to harness the full potential of AI. In November 2025, researchers at Google announced Project Suncatcher, which aims to scale machine learning computing in space in the future.
In January 2026, this area exploded. Kathleen Curlee, who studies the space economy at Georgetown University in Washington DC, told Nature magazine that everything took off. SpaceX, Elon Musk’s space technology company based in Starbase, Texas, shared plans to launch one million satellites to form an orbital data center. By comparison, there are currently about 15,000 satellites in low Earth orbit.
Not wanting to be left behind, the China Aerospace Science and Technology Corporation in Beijing joined the race at roughly the same time. Later, Blue Origin, Jeff Bezos’ space technology company based in Kent, Washington, filed an application for its own constellation.
Growing pressure on Earth
Technology companies are under increasing pressure over the location of their data centers on Earth. These facilities consume vast amounts of electricity – according to forecasts cited in the article, data centers will use twice as much energy by 2030 due to artificial intelligence. They also require significant amounts of water for cooling, which in drought‑affected regions has become a political issue.
In March 2026, the administration of US President Donald Trump introduced the Ratepayer Protection Pledge. Artificial intelligence companies such as Google, OpenAI and Musk’s xAI signed the pledge, committing to build their own power infrastructure or to buy the electricity needed for their data centers, to prevent shifting costs onto US residents. It is a non‑binding agreement, but by rolling it out before the November mid‑term elections, Trump made it clear that data centers are a political issue that could affect voters.
Putting orbital data centers in space could theoretically solve many of these problems. Satellite constellations could use solar energy instead of drawing electricity from the Earth’s grid. Cooling would be provided by the naturally cold environment of space, rather than the planet’s water sources. Moreover, satellites would not occupy large tracts of land.
Cooling in a vacuum
To make these projects successful, several engineering hurdles must be overcome. One is ensuring proper cooling of satellite electronics. Although space is much colder than Earth, it is also a vacuum. This means that the extreme heat generated mainly by AI systems is unlikely to dissipate on its own. In a vacuum, there is no convection or conduction of heat through air; the only way to release heat is by infrared radiation, which is far less efficient.
Igor Bargatin, a mechanical engineer at the University of Pennsylvania in Philadelphia, notes that cooling technologies for gadgets in space already exist, such as the radiators on the International Space Station. However, these are probably too heavy – and therefore too expensive to transport – to be used for orbital data centers. Radiators require large surface areas to radiate heat, and launching them into orbit incurs enormous costs. For comparison, launching one kilogram of payload into low Earth orbit using a Falcon 9 rocket costs about $2,500‑3,000. A million satellites with an average mass of 200 kilograms each would mean a transport cost of roughly $500 billion – an unrealistic amount even for the largest companies.
Solar power in orbit
The concept assumes that the satellite constellations would use solar energy instead of drawing power from the terrestrial electricity grid. In orbit, above the atmosphere, solar panels operate more efficiently – there are no clouds, dust or sunset (though in low Earth orbit a satellite enters Earth’s shadow every 90 minutes, so batteries are needed to survive eclipses). Average solar irradiance in orbit is about 1,360 watts per square metre, while on the Earth’s surface it typically ranges from 250 to 1,000 watts, depending on weather.
However, even with efficient solar panels, a satellite with computing power comparable to an Earth‑based AI server would need huge arrays. Graphics processing units (GPUs) used for training AI models, such as the NVIDIA H100, draw about 700 watts each. Thousands of such chips on a single satellite would generate megawatts of heat. Solar panels with an area of several hundred square metres would be needed to power just one satellite, making it large and expensive.
Costs and logistics
Companies are lobbying for launch permits and are pressing to make space data centers a reality within the next few years. However, scientists who spoke to Nature magazine believe that implementing this science‑fiction technology will take longer. First, current rockets cannot launch a million satellites in a short time. SpaceX, which leads in launch frequency, conducted a total of 140 successful missions in 2025 (according to data from the company’s website). Each Falcon 9 rocket can carry about 60 Starlink satellites at once. To place one million satellites into orbit would require more than 16,000 launches – more than 100 years at the current pace.
Blue Origin is just beginning regular launches of its New Glenn rocket, which can carry about 45 satellites at a time. China has reusable programmes (the Long March 9 rocket is in the testing phase), but so far it has not achieved SpaceX’s scale. Even the most optimistic scenarios assume it will take at least 15‑20 years to build an orbital data center.
Another issue is maintenance costs. Satellites in low Earth orbit degrade under cosmic radiation and residual atmospheric drag, which causes them to slowly descend. A typical Starlink satellite has a lifespan of about five years. For an orbital data center, replacing faulty units and replenishing the constellation would have to be done continuously, generating additional costs.
Concerns of local communities
Meanwhile, on Earth, local communities are not waiting for space‑based solutions. In Michigan, a township board voted in May 2026 to impose a one‑year moratorium on providing water to hyperscale data centers. The decision came after a technology company (unnamed) submitted plans for a facility that would consume millions of gallons of water per day for server cooling. Residents protested, arguing that in a region that has suffered droughts in recent years, priority should be given to drinking water supplies and irrigation.
Similar protests took place in Arizona, where in August 2025 the community gathered to oppose the installation of a data center by Amazon Web Services. In Virginia, home to the world’s largest concentration of data centers (so‑called Data Center Alley), residents complain about noise from cooling systems and rising electricity prices.
Ratepayer Protection Pledge
The Ratepayer Protection Pledge, announced in March 2026 by the Trump administration, is a response to these concerns. Companies that signed it committed to building their own energy infrastructure (e.g., solar farms, wind farms or small modular reactors) or to buying electricity at market rates without shifting costs onto retail customers. In return, the administration promised faster permitting procedures for data center construction.
Signatories include Google, OpenAI (creator of ChatGPT) and xAI – Elon Musk’s artificial intelligence company, which is separate from SpaceX. Some smaller cloud‑service providers did not sign, arguing that they cannot afford investments in their own energy sources. The agreement is non‑binding but has political significance in the context of the upcoming November 2026 mid‑term elections.
Can space save AI?
The idea of putting data centers in space is not detached from reality, but for now it remains in the realm of theory and plans. SpaceX, Blue Origin and Chinese space companies are competing to see who will build a working constellation first. Scientists are sceptical, pointing to technical (cooling, power supply) and economic (launch and maintenance costs) barriers. Kathleen Curlee of Georgetown University told Nature that while the idea is fascinating, she does not expect a fully functional orbital data center to be built in the next 5‑10 years.
