Space and AI: Could data centres move into orbit?
Artificial intelligence (AI) already handles more data than our planet can support. As they race for global dominance, AI market leaders are looking beyond Earth at the possibility of putting vast new data centres into orbit.
In 2026 alone, operators have filed plans for tens of thousands of satellites – and in one case a million – that could boost space-based computing and data storage. Satellites of unprecedented size are on the drawing board, alongside concepts for multi-satellite distributed computing.
“Orbital facilities could diversify where computing takes place,” said Jermaine Francisco Gutierrez, Research Fellow at the European Space Policy Institute (ESPI).
Increasingly intelligent satellites will need computing power, while orbital data centres will depend on rapid transmission and complex radio-frequency coordination.
The latest episode of Space Connect from the International Telecommunication Union (ITU) illustrates why and how data could go orbital, along with what the shift could mean for people on Earth.
Why move “compute” to space?
Motives vary, Gutierrez suggested. For Europe, this is “very much a sovereignty play” to reduce reliance on external cloud providers. China is focused on easing bandwidth limits in its satellite networks. US technology companies, meanwhile, aim to extend their head start in cloud services for decades to come.
Some compute capabilities already exist in space. Data storage and edge computing, where data is processed close to where it is generated rather than sent elsewhere first, already operate in orbit.
“Security and defence are the drivers for the moment,” said Ian Carnelli, Head of the Systems Department at the European Space Agency (ESA).
Links between space systems could eventually allow AI to assign tasks directly to satellites in orbit. “This is going to be a revolution,” Carnelli said.
Space-based training of large language models – the technology behind generative AI tools such as today’s chatbots – remains a more distant prospect.
A clearer near-term opportunity lies in processing Earth-observation data onboard. A single pass by a satellite such as Sentinel-2 generates terabytes of data, much of which currently goes unprocessed.
When the network becomes the bottleneck
Orbital data centres face two very different engineering barriers, said Brandon Karpf, who leads international public-private partnerships at NTT, the Japanese telecommunications and technology group.
A data centre could take the form of a single large orbital platform or distribute its computing across many smaller satellites.
A single large platform runs into heat and mass constraints. Distributing computing across smaller satellites creates another problem: links between them may carry about 100 gigabits per second, compared with only around 7 terabits per second between chips on Earth.
Karpf compares the scale of current terrestrial and near-future space computing as “garden hose versus municipal fire hydrant.”
Of five potential uses for orbital computing he modelled, two failed on physics and two on economics. The remaining case was a sovereign cloud, designed to keep data under the jurisdiction of a chosen country, “because it sells jurisdiction rather than speed or price.”
“Compute is abundant,” Karpf said. “Bandwidth is scarce, contested, and governed, and those three things together are what make a market.”
Demand for bandwidth keeps growing, particularly with robots and autonomous systems transmitting video and sensor data. This makes decisions about who can use which radio frequencies a crucial policy decision.
Choices of the best business case will often come down to “a spectrum decision” Karpf said, acknowledging ITU’s role in coordinating spectrum and orbit usage.
Rules for computing beyond Earth
Orbital data centres will also need to coordinate their radio-frequency use and associated orbits through ITU. Other regulatory questions relate to cybersecurity, intellectual property, AI governance and control over data.
“No jurisdiction on Earth has adopted a dedicated regulatory framework for orbital data centres,” said Julie Kearney, Co-Chair of the Space Exploration and Innovation Practice and partner in the telecom group at global law firm DLA Piper.
Dialogue must continue through ITU, as well as the United Nations Committee on the Peaceful Uses of Outer Space (COPUOS) and other international organizations, to “make sure that we are developing responsibly,” Kearney said.
Climate and environmental impact is another concern surrounding terrestrial data centres. By 2030, their electricity consumption could be equivalent to Japan’s current electricity use, with AI accounting for much of the growth.
Moving computing infrastructure into orbit would transfer, rather than eliminate, environmental pressures. Rocket launches, orbital debris and atmospheric re-entry would take the place of terrestrial power generation, water consumption and land-use issues.
“Putting data centres in space does not resolve the artificial intelligence infrastructure problem,” agreed Narda M. Jones, founder of Acacia Hill Strategies.
Orbital data centres would link to ground stations and fibre connections, which still disrupt local ecosystems. But such projects can either benefit – or bypass – local schools, hospitals and businesses.
As Jones see it: “We have an opportunity to address these choices before the architecture hardens.”
AI becomes space infrastructure
AI could accelerate demand for orbital computing long before full-fledged data centres reach space.
Wang Jian, Director of Zhejiang Lab and founder of Alibaba Cloud, envisions satellites that process data and collaborate rather than functioning as isolated sensors. The goal is “really computing in space, not just a chip in space,” he said during ITU’s last AI for Good Global Summit.
Autonomy becomes increasingly important farther from Earth. AI enables spacecraft and robots to operate autonomously when communication delays make real-time control from Earth impractical.
Bilge Demirköz, an astro-particle physicist at Middle East Technical University and member of the UN’s Independent International Scientific Panel on AI, pointed to the autonomous Ingenuity helicopter flown on Mars between 2021 and 2024.
“If you’re waiting for instructions from Planet Earth, you’re not going to be able to fly a helicopter,” she said.
Space-based computing could also ease mounting AI-related resource demand.
“The sun has unlimited energy,” commented Stephanie Bednarek, Vice President of Commercial Sales at SpaceX. “We don’t need water, land. And like we say, jokingly, it is always sunny in space.”
Play the episodes:
The series continues
ITU’s Space Connect series explores the space ecosystem and pathways to future communications, with upcoming sessions set to highlight policy challenges, emerging space innovations and more.
The next episode, a review of Constellations and Radio Astronomy (6 November 2026), explores how expanding satellite networks affect advanced telescopes and what this means for science. Learn more and register.
Learn more about the ITU Radiocommunication Sector (ITU-R) and its Space Services Department.
Header image credit: ITU
