As AI drives a new wave of infrastructure demand, cloud, AI, and space companies are exploring whether some computing and storage workloads can move into orbit. Recent activity as reported by Via Satellite June 2 includes SpaceX filing for a constellation of up to 1 million satellites to create an orbital data center, Google exploring Tensor Processing Unit clusters in space, Starcloud planning an 88,000-satellite constellation, and companies such as Ramon.Space, Planet, Lonestar Data Holdings, and Edge Aerospace building or testing pieces of the enabling infrastructure.
The idea sounds futuristic, but the market pressure is immediate. US spending on data center infrastructure jumped nearly 70% between May 2023 and May 2024, according to the American Edge Project, while the Lawrence Berkeley National Laboratory has projected data center energy consumption could double or triple by 2028 and account for up to 12% of US electricity use.
Plus, communities in major data center markets are already pushing back against land use, power demand, and water consumption tied to hyperscale campuses.
That pressure is forcing the industry to test more radical options. Orbital data centers are one of them. The harder question for enterprise technology leaders is not whether compute can operate in space. Early demonstrations suggest it can. The real question is which workloads make sense there, when the economics work, and whether orbital infrastructure solves a business problem that terrestrial data centers cannot.
The Case for Leaving the Ground
The argument for orbital infrastructure starts with terrestrial data center constraints. Land is scarce in key markets, power is increasingly contested, and cooling requirements strain water resources. Space changes that equation by offering a power source, physical room, and radiative cooling instead of water-based cooling systems.
Austin Litteral, director at aerospace and emerging technology investment firm Alpha Funds, framed the issue around those three constraints: land, power, and water. In space, he told Via, land is effectively vast and solar power is available.
Ramon.Space CEO Avi Shabtai compared it to the rise of cloud mega-infrastructure two decades ago. Even so, he acknowledged scaling orbital data centers depends on solving difficult engineering problems, including energy generation and storage, thermal management, radiation exposure, transportation, and in-orbit maintenance.
Satellite imagery and Earth observation company Planet is already working on early demonstrations. It has flown Nvidia GPUs on Pelican satellites to support on-orbit AI processing and is collaborating with Google on a phased orbital compute demonstration scheduled for early 2027 using Planet’s Owl next-generation satellite constellation. James Mason, Planet’s chief space officer, said the first phase is focused on risk reduction for Google, while also helping Planet fund higher-power satellite development for its own Earth observation work.
Analysis
What this means: AI infrastructure strategy is a physical-location question. As AI increases pressure on land, energy, cooling, and grid capacity, enterprise infrastructure planning will extend beyond traditional data center regions. ERP leaders may not move core systems into orbit, but they will face a broader set of infrastructure options shaped by power availability, sovereignty, resilience, and regulatory exposure.
Is Orbital Storage More Realistic?
The near-term business case for orbital data centers is not frontier AI training. Several industry voices argue that storage, edge compute, and sovereign data use cases are more credible starting points.
Stephen Eisele, president of Lonestar, said the near-term “killer application” is storage because it uses far less power than compute-heavy workloads and can be distributed across many satellites for resilience. Litteral made a similar point, saying near-term revenue is more likely in edge compute, where satellites already on orbit process data closer to where it is created.
Edge Aerospace is taking an enabling-layer approach. The Luxembourg-based company is developing low-SWaP computing systems, hybrid onboard computers, and network switches rather than trying to put GPU-heavy servers in space immediately. CEO Jaroslaw Jaworski said the company is building products it can sell today while moving toward orbital data centers over time.
Still, the technical gap remains significant. Brandon Karpf, who leads international security partnerships at NTT, analyzed several orbital compute business models and found only one convincing near-term case: sovereign cloud. His skepticism centers on communications bandwidth. Current optical satellite links deliver around 100 gigabits per second, with next-generation links potentially reaching 400 Gbps. By contrast, Nvidia connects a single GPU into a training cluster at approximately 7.2 terabits per second, before accounting for the tens of thousands of GPUs needed for frontier AI training.
Public cloud workloads face a different problem: elasticity. Fixed hardware launched into orbit cannot easily scale up or down the way cloud infrastructure does on Earth. Content delivery is also difficult to justify because terrestrial content delivery networks are already low-cost, high-performance, and embedded close to users.
Analysis
What this means: Orbital compute will test workload discipline. AI training, public cloud, content delivery, edge inference, and sovereign storage have very different technical and economic requirements. Enterprise architects should watch which workloads prove viable first, because the market will be shaped less by the idea of “data centers in space” and more by whether specific use cases can overcome bandwidth, latency, maintenance, launch, and governance constraints.
Sponsor Industry‑Grade Research
The Clearest Enterprise Case Is Sovereign Cloud
Sovereign data storage may be the strongest early enterprise and government use case.
Governments and regulated organizations increasingly want data infrastructure that reduces exposure to foreign legal access, physical attack, and geopolitical risk. A satellite-based storage vault can remain under a nation’s jurisdiction while being physically unreachable to adversaries on the ground.
Lonestar’s Chris Stott argued the global regulatory environment is tightening around where data can be processed and stored. Karpf also sees sovereign cloud as the most defensible business model because it competes on jurisdictional control rather than raw compute performance.
Orbital infrastructure is unlikely to replace mainstream enterprise cloud in the near term. But it could become relevant for specific categories of data, especially where sovereignty, resilience, defense, intelligence, or disaster recovery requirements justify higher complexity.
Those use cases align more closely with government, defense, regulated industry, and critical infrastructure than with general-purpose ERP workloads. But the direction of travel is notable. Data infrastructure strategy is becoming more geopolitical, and the physical location of compute and storage is becoming part of enterprise risk management.
Orbital data centers are not a near-term answer to every AI infrastructure constraint. They are a serious emerging option for specific workloads where terrestrial infrastructure is constrained by sovereignty, resilience, power, location, or access. The companies that focus on those use cases first are more likely to turn orbital compute into a viable market.
Analysis
What this means: Sovereign storage is the most credible near-term orbital use case. The strongest enterprise argument for orbital infrastructure centers on data sovereignty and resilience rather than general cloud replacement. Government, defense, critical infrastructure, and regulated industries are the most likely early adopters because orbital storage can address jurisdictional and physical-security concerns that terrestrial data centers cannot fully solve.
