Insider Brief
- Orbital AI data centers could ease pressure on electricity grids, water supplies and land, but would create new legal, cybersecurity and environmental risks.
- Current laws cover launches, communications and remote sensing but provide no clear U.S. authorization process for computing operations conducted in orbit.
- Falling launch costs and better thermal management could improve feasibility, although radiation, maintenance, cyberattacks and orbital debris remain major obstacles.
- Image: NASA
Orbital data centers could ease some of artificial intelligence’s pressure on energy grids, water supplies and land, but a new legal analysis warns that moving computing infrastructure into space would create difficult questions over cybersecurity, data ownership, liability and orbital debris.
The working paper, “Governing Orbital Data Centers in the New Space Race,” examines how existing international space law and U.S. regulations would apply to satellites used to store data, train AI models or run AI applications. Researchers Lauren E. Diaz and Vicenç Feliú conclude that orbital data centers would not operate beyond the reach of the law, despite being located outside any nation’s territory.
The larger problem is that today’s rules were not designed for privately operated computing constellations that could eventually contain thousands or even hundreds of thousands of satellites.
International space treaties establish broad principles for activity in orbit, while U.S. agencies regulate functions such as launches, reentry, radio communications and remote sensing. No single U.S. agency, however, has clear authority over the computing operations of a data center once it reaches orbit, according to the paper.
That gap could become more important as companies move beyond small experiments. Axiom Space has tested commercial data processing aboard the International Space Station and deployed dedicated orbital computing nodes. Google’s Project Suncatcher is exploring constellations of solar-powered satellites equipped with AI chips and optical links. SpaceX, Blue Origin and a growing group of startups have also outlined plans for space-based computing systems.
Those projects remain far smaller than the vast data centers powering leading AI systems on Earth. Still, the researchers argue that the arrival of prototypes, formal regulatory filings and commercial services makes orbital computing a present policy issue rather than a distant science-fiction scenario.
Why Companies Want AI Data Centers in Space
The push toward orbit begins with the physical demands of AI. Data centers consumed about 415 terawatt-hours of electricity worldwide in 2024, or roughly 1.5% of global electricity use, according to International Energy Agency figures cited in the paper. The agency projects consumption could rise to approximately 945 terawatt-hours by 2030, with AI accounting for much of the increase.
Large terrestrial data centers also require land, grid connections and substantial cooling systems. Some facilities consume millions of gallons of water a day, contributing to opposition in communities where water or electricity supplies are already strained.
Space appears to offer an alternative because satellites in suitable orbits could receive sunlight for most of the day without interference from clouds. Solar panels could generate electricity for computing, while the systems would avoid using terrestrial water or occupying valuable land.
Orbital computing could be particularly useful for information already produced in space. Earth-observation satellites, for example, can generate large volumes of images that must be transmitted to the ground for processing. Running AI models in orbit could allow a satellite network to identify useful information first and send only the results, reducing communications requirements and delays.
The concept becomes more difficult when companies try to recreate hyperscale data centers in orbit. Space is a vacuum, which means heat cannot be removed through air or ordinary water-cooling systems. Satellites must radiate heat away through large surfaces, making thermal management one of the most important engineering constraints.
Computer equipment must also withstand radiation, vibration during launch and years without direct physical maintenance. Repairs that are routine on Earth could require robotic servicing, replacement satellites or complete deorbiting.
Launch economics pose an even larger obstacle. Every processor, solar panel, radiator and communications component must be carried into orbit. The paper notes that orbital data centers are unlikely to compete with terrestrial facilities unless reusable heavy-lift rockets reduce launch costs substantially and operate at a much higher frequency.
The researchers therefore caution against treating orbital data centers as an established answer to AI’s energy problem. Current systems demonstrate that computing can take place in orbit, but they do not prove that space can support data-center operations at terrestrial scale or cost.
Cybersecurity, Data Privacy and AI Oversight
Cybersecurity represents one of the most immediate risks, according to Diaz and Feliú.
Orbital data centers would depend on communications links for commands, software updates and the transfer of information. Those links could be intercepted, jammed or spoofed. A successful attack could allow an adversary to steal data, alter an AI model, disrupt services or gain control of a satellite.
The physical consequences could extend beyond the loss of computing capacity. An attacker that compromised a satellite’s command system might alter its orbit, deliberately deorbit it or create a collision with another spacecraft.
Responding to such an attack would be harder than responding to a breach at a terrestrial data center. Technicians could not walk into the facility, disconnect compromised equipment or inspect it directly. Investigation and recovery would depend on remote access to systems that an attacker might already control.
Existing cybersecurity frameworks offer a starting point, but they were not developed specifically for orbital computing. The researchers recommend stronger requirements for encrypted command links, authenticated software updates, supply-chain security, anomaly detection and operations during communications outages.
Data privacy presents another unresolved issue. Outer space is not sovereign territory, but a satellite remains under the jurisdiction and control of the country where it is registered. Data stored aboard a U.S.-registered satellite operated by a U.S. company would therefore probably remain subject to U.S. law.
That means a company could not necessarily avoid privacy rules, court orders or law-enforcement demands by moving its servers into orbit. The researchers compare a registered satellite to a nationally flagged ship operating in international waters. Its physical location does not erase the legal authority of the flag state.
Other countries could also assert jurisdiction based on where customers live or where an AI system’s output is used. The European Union’s AI Act, for example, can cover systems placed on the EU market or producing results used within the bloc, regardless of where the provider or computing hardware is located.
That could leave an orbital data center answering to several legal systems at once. A satellite might be registered in one country, operated by a company incorporated in another and process personal information belonging to people in dozens of others.
The paper warns that some operators could nevertheless portray space as a regulatory haven. The researchers argue that governments should explicitly establish that orbital processing does not exempt AI companies from privacy, consumer-protection, cybersecurity or AI-safety rules.
Oversight could still be difficult with regulators possessing the authority to audit a powerful AI model, but examining hardware, training records and system logs becomes more complicated when the equipment is hundreds of miles above Earth. Communications interruptions could also limit the ability for humans to supervise AI systems controlling spacecraft or making collision-avoidance decisions.
Liability and the Growing Orbital Debris Problem
International law places responsibility for private space activities on governments, making states responsible for authorizing and supervising companies under their jurisdiction.
The Liability Convention generally makes a launching state absolutely liable for damage that its space object causes on Earth or to aircraft. Damage to another spacecraft is governed by a fault-based standard.
Orbital data centers would complicate both rules. If hackers took control of a computing satellite and caused a collision, investigators would have to determine whether fault rested with the attacker, the satellite operator, its cybersecurity suppliers or the government responsible for supervising the mission.
The researchers identify orbital debris as another major threat to the idea’s long-term viability. Low Earth orbit is already crowded with operating satellites, discarded hardware and fragments from previous collisions and destructive weapons tests.
Adding large computing constellations would increase the number of close approaches and collision-avoidance maneuvers. A collision could create thousands of fragments, some of which could strike other satellites and produce still more debris. In the most extreme case, that chain reaction could make parts of low Earth orbit unsafe for future operations.
Existing international debris guidelines are largely voluntary. Some countries have adopted national rules, including requirements for satellites to leave orbit after their missions, but the standards are not consistent worldwide.
The paper recommends stronger international requirements for satellite maneuverability, tracking, collision avoidance and end-of-life disposal. Governments could also require operators to post financial bonds that would be forfeited if they failed to remove their satellites, with the money supporting debris-cleanup efforts.
The environmental trade-off extends to Earth. Space-based solar power and radiative cooling could reduce electricity, water and land demands from terrestrial data centers. But manufacturing and launching vast constellations would create emissions and introduce particles into the upper atmosphere. Large satellite networks could also increase light pollution and radio interference for astronomers.
The net environmental benefit would depend on the number and lifetime of the satellites, the frequency and emissions of launches, the sources of electricity displaced on Earth and the ability of operators to prevent failed equipment from becoming debris.
Diaz and Feliú recommend a clearer U.S. authorization process for computing activities in orbit, cybersecurity requirements tailored to spacecraft and international standards for managing large constellations. They also call for disclosure of major cyber incidents, disposal performance, collision statistics and basic mission purposes.
The analysis’s limitations include that this is a legal working paper examining a fast-moving industry, not evidence that hyperscale AI data centers in space are technically or economically ready. Many of the projects it discusses remain prototypes, research programs or regulatory proposals.
