Powering AI from space, at scale

(Press-News.org) Penn Engineers have developed a novel design for solar-powered data centers that will orbit the earth and could realistically scale to meet the growing demand for AI computing while reducing the environmental impact of data centers. 

Reminiscent of a leafy plant, with multiple, hardware-containing stems connected to branching, leaf-like solar panels, the design leverages decades of research on “tethers,” rope-like cables that naturally orient themselves under the competing forces of gravity and centrifugal motion. This architecture could scale to the thousands of computing nodes needed to replicate the power of terrestrial data centers, at least for AI inference, the process of querying tools like ChatGPT after their training concludes.

Unlike prior designs, which typically require constant adjustments to keep solar panels pointed toward the sun, the new system is largely passive, its orientation maintained by natural forces acting on objects in orbit. By relying on these stabilizing effects, the design reduces weight, power consumption and overall complexity, making large-scale deployment more feasible.

“This is the first design that prioritizes passive orientation at this scale,” says Igor Bargatin, Associate Professor in Mechanical Engineering and Applied Mechanics (MEAM) and the senior author of a paper describing the system, presented at the 2026 American Institute of Aeronautics and Astronautics (AIAA) SciTech Forum. “Because the design relies on tethers — an existing, well-studied technology — we can realistically think about scaling orbital data centers to the size needed to meaningfully reduce the energy and water demands of data centers on Earth.”

Putting AI in Orbit

In recent months, startups, public companies and even governments have proposed designs for solar-powered orbital data centers. The promise is straightforward: placing data centers in space could reduce their reliance on Earth’s increasingly strained electricity grids and water supplies.

“The problem is that these designs are challenging to scale,” says Bargatin. “If you rely on constellations of individual satellites flying independently, you would need millions of them to make a real difference.”

Other proposals envision enormous rigid structures assembled robotically in orbit. In theory, these systems could host large amounts of computing power, but their size and structural complexity place them beyond current manufacturing and deployment capabilities.

By contrast, the new design occupies a middle ground: ambitious enough to make an impact, yet simple enough to plausibly deploy using technology that already exists and is well understood, like tethers, which have been studied and tested in space for decades.

The Power of Tethers 

Proposed in the early days of the space age, tethers are essentially long, flexible cables that behave in unique ways once in orbit. Pulled taut by the competing forces acting on orbiting objects — Earth’s gravity and the centrifuge-like effect of orbital motion — tethers naturally align themselves, with one end pulled earthward and the other extending towards space.

In the Penn design, thousands of identical computing nodes would be connected along a tether, forming a long, vertical chain in orbit. Each node would carry computer chips, solar panels and cooling hardware, creating a modular structure. “Just as you can keep adding beads to form a longer necklace,” says Bargatin, “you can scale the tethers by adding nodes.” 

Sunlight itself would keep the solar panels correctly aligned. The gentle but constant pressure exerted by the sun’s rays acts much like wind on a weather vane, keeping the panels oriented without motors or thrusters. “We’re using sunlight not just as a power source, but as part of the control system,” says Bargatin. “Solar pressure is very small, but by using thin-film materials and slightly angling the panels toward the computer elements, we can leverage that pressure to keep the system pointed in the right direction.”

In simulations described in the paper, a single tethered system could stretch for several or even tens of kilometers, hosting thousands of computing nodes and supporting up to 20 megawatts of computing power, equivalent to a medium-sized data center on Earth. Data processed onboard would be transmitted using laser-based optical links, a technology already used to relay information between satellites in orbit. 

Even though the time it takes to transmit training data to and from Earth would likely preclude training AI in space, much of the growth in AI demand, as Bargatin notes, will be in queries to already trained AI systems — precisely the task the tether-based system is designed to support. “Imagine a belt of these systems encircling the planet,” says Bargatin. “Instead of one massive data center, you’d have many modular ones working together, powered continuously by sunlight.”

Robust to Impacts

Any large structure operating in orbit must contend with constant impacts from micrometeoroids, tiny fragments of debris and dust traveling at extreme speeds. “It’s not a matter of preventing impacts,” says Jordan Raney, Associate Professor in MEAM and a co-author of the paper. “The real question is how the system responds when they happen.”

Raney and Dengge “Grace” Jin, a doctoral student in MEAM, used computer simulations to model how micrometeoroid impacts would affect the tethered structure over time. Rather than focusing on a single collision, the team examined the cumulative effects of many impacts distributed across the system.

The results suggest that the tethered design is naturally resilient. When a micrometeoroid strikes part of the structure, the impact can cause a brief wobble or rotation, but that motion spreads along the length of the tether and gradually dissipates. “It’s a bit like a wind chime,” Raney says. “If you disturb the structure, eventually the motion dies down naturally. We had to understand how long that process would take, to be sure that the data center would be stable even when hit by multiple objects.”

In the wide range of scenarios the researchers simulated, the tether-based system deviated from its optimal orientation by only a few degrees. “Each node is supported by multiple tethers,” Raney notes. “So even if an impact severed a tether, the system would continue to function.” 

Future Directions

Unlike terrestrial data centers, which rely on air or liquid cooling, space-based systems can shed heat only by slowly radiating it away. While the proposed system includes radiators to shed waste heat, Bargatin hopes to improve their design, with the goal of developing lightweight, durable radiators capable of dissipating the heat of sustained computing loads in space.

Next, the researchers hope to move beyond simulations, building and testing their design as a small prototype, with a limited number of nodes. “Much of the growth in AI isn’t coming from training new models, but from running them over and over again,” says Bargatin. “If we can support that inference in space, it opens up a new path for scaling AI with less impact on Earth.”

This study was conducted at the University of Pennsylvania School of Engineering and Applied Science (Penn Engineering) and was supported by Penn Engineering. Additional co-authors include Zaini Alansari of Penn Engineering. 

END