The issue of building datacenters has garnered a lot of attention lately, with good reason. With dynamics including skyrocketing demand for AI compute, land for construction at a premium, and proposed sites bumping up against the NIMBY (Not In My Back Yard) effect, it seems logical to look to the skies and put datacenters in space, right?
Companies including Starlink, Amazon, Blue Origin, and SpaceX think so. Already, they and others have been deploying and testing orbital datacenter hardware, so the answer is, perhaps.
But it’s a tricky proposition.
There’s no question about the need. By the end of the decade, global annual datacenter power consumption is expected to nearly double to nearly 1,000 terawatt-hours, according to an estimate by the International Energy Agency.
Further, moving datacenters into space looks better when considering how power demand is increasing, energy costs are rising, and the effects of scalability on Earth.
“Approximately 95% of solar energy is lost by the time it reaches Earth, and computing in orbit could capture far greater efficiency,” according to the National Security Data and Policy Institute at the University of Virginia. With China already planning to integrate orbital computing into its military and intelligence strategy, “Space-based AI infrastructure is not some science fiction plot, but an emergent domain for geopolitical competition,” the Institute said.
However, orbital datacenters are fraught with challenges. Experts underline the importance of understanding what they are and what they are not.
“The vision for orbital datacenters is not hyperscale facilities in space. Rather, it’s a modular, networked layer of compute satellites designed for workloads where orbit provides structural advantages,” said Sean McDevitt, a partner at management consulting firm Arthur D. Little.
Those advantages include near-continuous solar exposure (in a low orbit close to 1,400 km), a passive thermal environment, proximity to space-generated data, and/or geopolitical resilience, McDevitt said. “Additional benefits include the scale effect of similar modules, mass optimization, and compatibility with a launcher’s capacity and volume.”
Orbital datacenters are being explored because the bottlenecks being seen in AI infrastructure have moved beyond server racks, McDevitt added. “The question now is whether we can build enough, power [a datacenter], permit it, and bring it online fast enough. Space looks like an alternative to do that.”
Some say orbital datacenters make sense now because the cost of hardware has come down. “With some solid but not necessarily heroic engineering, the cost of an orbital datacenter could be as low as three times that of the comparable terrestrial one,” according to IEEE Spectrum.
That assessment calculated the total cost of a datacenter satellite network at around $51 billion, including launch and five years of operational expenses. In comparison, an equivalent terrestrial system would cost about $16 billion over the same period.
Tom Coughlin, an IEEE life fellow, said the decreasing price of space transportation is due to commercial spacecraft participation with reusable components and regular standardized launch schedules. “Increased scale brings cost reductions,” he noted. “This decreases the cost of launching things into space which, in turn, lowers the cost of any activity in space.”
There is also a growing amount of activity in space, he said. Initially concentrated in low Earth orbit, it could eventually be further out to include cis-lunar orbits, the lunar surface, and elsewhere in the solar system.
Coughlin issued a caveat: “These activities, especially the industrialization of outer space, will greatly depend on robotics,” meaning physical AI, “and require local datacenter resources to enable them. That is the ultimate reason for having datacenters in space.”
The benefits of orbital datacenters are to enable the industrialization of outer space, which will be highly dependent upon automation and access to local computing resources, according to Coughlin. “Because of the dangers and costs for humans working in space, much of this development will be done using automation, requiring local processing to support real-time activities.”
The big challenges are radiation sensitivity of many electronic components, and heat dissipation without access to convection. “Heat dissipation in space must be done using radiation,” Coughlin said. “That is a major reason why space datacenters are largely seen as smaller computing systems in a networked constellation, since that maximizes the surface area to volume of the space-based datacenter. Dealing with errors due to radiation will lead to higher amounts of error correction and the use of components that are less sensitive to radiation.”
Experts say orbital datacenters are not meant to replace those on Earth, but to handle specialized workloads. Orbital systems would handle asynchronous, energy-intensive workloads, while Earth-based datacenters remain dominant in real-time computing, according to commercial real estate and investment firm JLL.
“If the economics are compelling and you’re able to get around the risk of debris management . . . which needs to be overcome, and launch costs fall below threshold,” then orbital datacenters without huge latency requirements are not a bad idea, said Daniel Thorpe, head of datacenter research, EMEA, at JLL and co-author of the report, Data Centers in Space: The forces pushing and pulling datacenter infrastructure above the cloud.
However, anything linked to inference, Thorpe said, will require a datacenter closer to the end user or customer. AI training is a good fit because it is not latency sensitive, as well as anything requiring drone technology, he said.
McDevitt said orbital datacenters should not be considered as replacements of terrestrial infrastructure, but rather as a complementary layer for specific workloads where space offers structural advantages.
Other use cases could include activities such as GPS, weather observation, communications, and scientific work, said Coughlin. “But I think the biggest beneficiary will be initially for the creation of manufacturing facilities in [low Earth orbit] or higher orbits, mining the Moon for various resources, and other industrial activities.”
McDevitt stressed that orbital datacenters will not be cheaper than those on Earth. And there are other hurdles to overcome. “There’s a regulatory component to this that people haven’t talked about,” he said. “Governance, rights, and sovereign rights of a nation—when you start putting stuff in space, there’s a regulatory dynamic to this people aren’t paying attention to.”
For example, there are a wide range of regulatory and geopolitical issues, such as spectrum allocation, orbital traffic management, debris rules, export controls, national security concerns, cyber governance, and questions over who can operate strategically important compute infrastructure in orbit, McDevitt said. “Very few people are talking about this now,” given that the market is in the nascent stage.
There is also a sovereignty angle: If orbital compute becomes part of critical digital infrastructure, “Governments will care deeply about control, jurisdiction, and dependence,” McDevitt said. “None of this may stop adoption, but even if/when proven commercially viable, these types of things could slow scaling and fragment the market by region and alliance bloc.”
JLL’s Thorpe said that in addition to obvious considerations about maintaining a datacenter in space, it won’t be easy to send an engineer into orbit if an issue arises in such a datacenter. There also are challenges like hardware obsolescence and the need to upgrade chips nearly every year, maintenance, launch costs, and congestion. Already, there are about 70,000 satellites in operation, half of which are in low orbit, and there is an issue with space debris.
“There is a large amount of untracked debris. There is a collision risk. On top of that . . . where does the liability sit” if a datacenter in space ends up destroying another satellite. “Who owns that liability and what does the insurance look like?” Those are key challenges to watch, Thorpe said.
Meanwhile, the race to put datacenters in space is rocketing forward. Right now, SpaceX is leading the charge, ADL’s McDevitt said. Coughlin the IEEE fellow, said we’ll see the underpinning of an orbital datacenter “this year, at some level. However, I think the major build-out of datacenters in space will happen over the next five years.”
Esther Shein is a freelance technology and business writer based in the Boston area.
Submit an Article to CACM
CACM welcomes unsolicited submissions on topics of relevance and value to the computing community.
You Just Read
Datacenters in Space
© 2026 ACM 0001-0782/26/7