The rapid advancement of large language models (LLMs) is fueling a global data center boom and driving a surge in energy demand. But the electricity required to power data centers is straining the grid, pushing infrastructure operators to search for alternative sources of power. Some are even looking beyond Earth.

One company that’s looking to the stars for energy is Orbital Inc. In mid-April, the Los Angeles–based startup emerged from stealth and announced plans to build space data centers. Backed by Andreessen Horowitz (A16z), Orbital is designing infrastructure for AI inference, where trained models generate outputs. Much like other companies advocating for space-based data centers, Orbital is banking on the “free” energy generated by the sun to power compute for workloads such as chatbots and agents, sidestepping terrestrial energy constraints.

“There simply isn’t enough capacity here [on Earth], and the only way is up,” says Euwyn Poon, Orbital’s founder and CEO. “There’s actually abundant solar energy that’s not being harnessed.”

Orbital’s vision is a mesh constellation of small satellites in low Earth orbit. Each satellite would be equipped with a GPU server rack powered by solar panels roughly the size of a tennis court, plus radiative cooling panels of comparable size. The long-term goal is up to 10,000 fridge-sized satellites—each with 100 kilowatts of power—forming a distributed cloud, similar to SpaceX’s proposed AI Sat Mini.

Orbital’s first test will come in 2027, when it plans to launch a prototype satellite aboard a SpaceX Falcon 9 to validate its GPU operations in orbit and run commercial inference workloads. Another company, Starcloud, has already run a similar test last year. Orbital’s differentiator is their plans to match the solution with a problem: Small satellites equipped to run inference workloads specifically could benefit from lower launch costs. However, they face the same difficulties as other space data center hopefuls. Every watt of “free” energy must be dissipated as heat via large radiative coolers; radiation in low Earth orbit degrades compute equipment; and regular maintenance in space is difficult and costly.

Orbital’s inference focus

Poon says Orbital’s focus on a distributed network of smaller satellites designed to run inference workloads across independent GPU nodes, rather than large, tightly-coupled systems, makes the execution more feasible.

That idea shapes Orbital’s design. Training large AI models typically relies on tightly-coupled GPU clusters optimized for massive compute throughput. Inference workloads, by contrast, are generally less compute-intensive per request and can often run on smaller numbers of GPUs, making them easier to distribute across systems. Capping each satellite at roughly 100 kilowatts, Poon says, greatly simplifies the design. “It’s very simple,” Poon says, referring to the concept behind the satellites’ engineering. “Engineers would appreciate this.”

In Orbital’s design, a user request—like, say, asking ChatGPT to analyze a data set—is routed from a data center on Earth to a ground station, a terrestrial relay that connects satellites to the internet, then transmits the request to a satellite. Satellites communicate through optical interlinks, which use lasers to pass data between nodes. That routes the request to an available GPU, which processes the user’s query and generates the output before sending the result back through the network to the user. These links rely on ground stations that only communicate with satellites when they pass within range.

If the satellites are proven to work, Orbital is set on tapping “big model labs” as customers, including firms like OpenAI and Anthropic that run massive inference workloads. Orbital plans to serve them through direct API access for buying tokens and enterprise deals that shift inference demand into its network in space.

Engineering challenges

Poon recognizes that running data centers in space introduces major technical hurdles.

Radiation can strike GPUs and cause bit flips or other errors. Thermal management is also difficult. Without air, systems must rely on radiating heat into space rather than conventional cooling. Maintenance is another constraint, as satellites cannot be easily repaired or replaced if they malfunction in space. It’s why Poon says the test launch will be critical to identify and troubleshoot these issues. “Part of the mission is to figure out the unknowns,” he says.

Dr. Amit Verma, an electrical engineering professor at Texas A&M University–Kingsville, who researches semiconductor device modeling, raised similar concerns. Deploying thousands of satellites, Dr. Verma says, increases failure risk with limited repair options. He added that operational feasibility depends on the applications performed on the satellites. While some workloads like chatbots or algorithmic recommendations can tolerate added delays (data traveling to lower Earth orbit takes tens of milliseconds to return), others like real-time stock trading cannot.

“Outer space data centers that involve heavy use of AI-related processing certainly do need to overcome power and deployment and reliability issues to be meaningful,” Verma says.

Orbital plans to test extensively before launch. Poon says his company is exploring radiation hardening for GPUs and ammonia-based liquid cooling loops to transfer heat to external radiators. Reducing system weight is also top of mind to lower launch costs.

Even with these mitigations, the timeline is ambitious. In a Substack post on space data centers, Andrew Côté, an engineering physicist, predicts that space data centers won’t be operational for at least another 10 to 20 years. Orbital, however, expects to finalize the satellite designs by 2026, launch in 2027, and build a manufacturing facility in Los Angeles by 2028.

With the engineering challenges complex and the costs of launch high, the ability for Orbital’s satellite systems to operate reliably at scale remains an open question.

Despite those uncertainties, Poon remains laser focused on the long-term opportunity.

“I trust that our engineering efforts can start making progress towards solving these problems,” he says.