When orbital AI data centers become business reality

The reality of orbital data centers

An orbital data center is a compute facility hosted on a satellite or group of satellites in low-Earth orbit.

Rather than pulling power from the terrestrial grid like their Earth-bound counterparts, orbital data centers would generate their own power from solar arrays. While the approach is an emerging phenomenon, the underlying idea isn't new. Engineer Peter Glaser first proposed space-based solar power in a 1968 paper in Science decades before AI considered putting compute into orbit.

Orbital data centers aren't intended to be a near-term substitute for terrestrial hyperscale capacity, but rather a complement to it. The scale varies enormously depending on whose design is being described. Prototype satellites already in orbit carry a single AI accelerator chip, with a compute footprint roughly the size of a high-end server. Proposed full-scale designs go much further, with research architectures built around dense clusters of dozens of satellites flying in tight formation and regulatory filings describing constellations of tens of thousands to as many as one million satellites for gigawatt-class capacity.

The value proposition for orbital data centers is straightforward. Satellites in sun-synchronous orbits receive near-continuous sunlight, generating power via solar arrays without competing for grid capacity or local water supplies. A satellite doesn't require the multiyear local permitting process that's now delaying many terrestrial builds.

As a result, investment in orbital compute is accelerating even though the underlying economics remain unresolved.

Economics of orbital AI data centers

Cost is the primary obstacle to orbital data centers. The most detailed cost comparison between orbital and terrestrial data centers is shown in an open, interactive cost model built by Andrew McCalip, head of R&D at Varda Space Industries, which he updates as assumptions change. As of July 20, 2026, a 1 gigawatt facility for the first five years is roughly $51.1 billion for an orbital solar build versus $15.9 billion for a terrestrial equivalent -- driven mostly by launch and hardware costs.

SpaceX does not even need to be able to deploy cheaper than on Earth. It just needs to be faster.

In a June 17, 2026 paper, Google researchers calculated that at today's prices, orbital power costs about $14,700 per kilowatt per year versus $570 to $3,000 for a U.S.-based data center. If launch costs fall to $200 per kilogram, a level they call plausible by the mid-2030s, the orbital power cost would drop to around $810 per kilowatt per year, well within the range of terrestrial data center power costs.

Deployment speed, not absolute cost, is also seen as a deciding factor. "SpaceX does not even need to be able to deploy cheaper than on Earth. It just needs to be faster," said Patrick Bowen, CEO of optical processing unit developer Neurophos. With generator lead times now approaching a decade, he argued that speed itself has become the binding constraint. "We're nearing the point where even if GPUs were free, you can't turn them on," Bowen said.

Timing is another open question. More than 17,000 satellites are currently in orbit with more than half of them in low-Earth orbit. Operators managing that congestion must keep pace with chip development cycles that move far faster than satellite lifecycles, according to a research report on space-based data centers co-authored by Daniel Thorpe, director and head of EMEA data center research at JLL, a commercial real estate agency.

"Relocating general AI workloads into space depends on a long list of factors that haven't fallen into place yet: launch costs, the inability to service hardware once it's up there and chips becoming obsolete faster than the satellites carrying them," Thorpe explained.

Orbital AI data center deployment challenges

Survivability, servicing and data movement are among the challenges to deploying AI data centers in orbit, according to Sebastien Jean, CTO at NAND storage provider Phison Electronics, which partnered with Lonestar Data Holdings to deploy data center hardware on the moon.

"Every data center in space still depends on hardware that can survive launch vibration, radiation exposure, temperature swings and long periods with limited physical access," Jean said. Also, hardware failures can't be addressed once a satellite is in orbit. "That raises the bar for reliability, maintenance and lifecycle planning, because swapping out a CPU, GPU or SSD is much harder when you can't walk over to the server," Jean noted.

In addition, AI infrastructure is constrained not only by compute, but also by security, resilience and data movement. "Moving the system into orbit, changes nearly all those variables at once, the hardest one being bulk data movement to orbital infrastructure," Jean added.

Current orbital data center development

Despite those constraints, several companies with serious capital behind them have already moved from concept to hardware, including the following:

• Blue Origin. Amazon founder Jeff Bezos' space venture recently revealed its orbital data center plans. The company filed with the FCC in April 2026 for "Project Sunrise," a constellation of up to 51,600 satellites to operate as a data center in space in sun-synchronous orbit at altitudes between 500 and 1,800 km (~310 and ~1,118 miles).
• Google. The company announced Project Suncatcher in November 2025, calling it a "research moonshot" in the same category as the company's work on autonomous vehicles and quantum computing. The research effort tests Tensor Processing Units on satellites flying in tight formation. The company plans to launch two prototype satellites in early 2027.
• Lonestar Data Holdings. The company reported launching four data centers into space, including two to the moon. The data centers are single server configurations with limited compute and storage capacity. Its highest-profile mission so far, called Freedom, reached the lunar surface in February 2025 aboard Intuitive Machines' Athena lander, launched on a SpaceX Falcon 9 rocket and equipped with Phison's Pascari Enterprise SSDs. More recently, Lonestar signed a NASA Space Act Agreement in May 2026 to advance lunar data storage and resilient off-world compute infrastructure.
• SpaceX. The company filed with the FCC in January 2026 for a constellation of up to one million satellites for orbital AI compute. The plan targets deployment "as early as 2028," working toward a long-term goal of 100 gigawatts of orbital compute capacity per year, which SpaceX said would require thousands of launches and delivering roughly one million metric tons of hardware to orbit annually, according to its S-1 filing with the SEC ahead of its June 2026 IPO. The satellites would fly in sun-synchronous orbit and link back to Earth through Starlink. The same filing lists the initiative among SpaceX's principal risk factors, cautioning it "may not achieve commercial viability."
• Starcloud. The company launched the first satellite carrying an Nvidia H100 GPU in November 2025, then trained the first LLM in orbit the following month. "The AI revolution is colliding with the physical limits of our terrestrial energy grid," said CEO and co-founder Philip Johnston in a statement. "We are quickly running out of places to build new energy projects for data centers on Earth. By moving AI compute to space, we unlock access to unlimited solar power and completely remove the energy bottleneck." 

Milestones to watch

It's still relatively early days for orbital AI data centers. "The headlines create an enormous sense of imminence, but commercial reality is still some way off," JLL's Thorpe said. The JLL report cited the following four key "thresholds" for orbital data centers to move from demonstration to commercial viability:

Lower launch costs. The JLL report cites a McKinsey analysis of Starcloud's business model showing that below $500/km, space-based compute becomes structurally cheaper than terrestrial computing, even after accounting for hardware replacement and data transmission costs. SpaceX's Starship targets $200/km, down from Falcon 9's current $2,700/km.

When underwriters are willing to insure orbital collisions and debris, that's the signal that capital has concluded the risk is real but manageable.

Hyperscaler validation. Google's Suncatcher satellites, planned for launch in 2027, are built to prove that hardware can survive and AI training can work in orbit at real latency, throughput and cost. Google's existing enterprise cloud customers and service-level obligations lend real weight to its results.

Manageable insurance risks. JLL counts roughly 44,000 tracked objects in Earth orbit large enough to destroy a satellite. The Kessler Syndrome, a theoretical scenario in which collisions among objects create a cascading chain of new debris that cause further collisions, is a growing threat among the 17,000-plus satellites already in orbit. "When underwriters are willing to insure orbital collisions and debris, that's the signal that capital has concluded the risk is real but manageable," Thorpe noted.

Terrestrial infrastructures failing to meet demand. Worsening ground conditions, like persistent grid delays and local community opposition, can make orbital compute look better. That's the same pressure behind the IEA's warning that new transmission lines can take four to eight years to build.

Until those milestones are reached, orbital data centers aren't a near-term procurement consideration for enterprise technology buyers. "For enterprise buyers, the question isn't simply whether AI workloads can run in orbit," Phison's Jean noted, "it's whether they can run there at a cost, risk level and latency profile that's better than mature terrestrial options."

When those questions are answered, the two workloads that will be most feasible for orbital data centers from a business standpoint are space-generated data and secure, resilient storage of vital Earth-based data, according to Kevin O'Connell, CEO and founder of consultancy Space Economy Rising and former director of the U.S. Department of Commerce's Office of Space Commerce.

Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He has pulled Token Ring, configured NetWare and been known to compile his own Linux kernel. He consults with industry and media organizations on technology issues.