SpaceX, Blue Origin and Starcloud target orbital data centers for AI

SpaceX, Starcloud, Blue Origin and several other players are pushing the idea of orbital data centers for AI. The logic is straightforward: move computing infrastructure to sun-synchronous orbits, where solar panels remain idle far less often and energy is available much more stably than on Earth.

Who's Going to Orbit

The catalyst for discussion came from applications to the U.S. Federal Communications Commission.

According to published data, SpaceX requested permission for a constellation of one million satellites at altitudes of 500 to 2,000 kilometers. Starcloud filed an application for 88,000 devices in the 600–850 kilometer range and has already launched the first prototype with an Nvidia H100 processor into orbit. Blue Origin joined in March 2026 with a project for another 51,600 satellites.

The list doesn't end there. Google, together with Planet Labs, is working on a system of 81 satellites, with the launch of two demonstration units planned for early 2027. Aetherflux, in turn, announced the first node of Galactic Brain in the first quarter of 2027.

If you count only the largest American projects, we're already talking about approximately 1.14 million computing satellites—which moves the idea from the category of futurism into the realm of infrastructure plans for the coming years.

• SpaceX — up to 1,000,000 satellites at altitudes of 500–2,000 km
• Starcloud — 88,000 devices and the first orbital prototype with Nvidia H100
• Blue Origin — 51,600 satellites in a project revealed in March 2026
• Google and Planet Labs — 81 satellites and two demo launches in early 2027
• Aetherflux — the first node of Galactic Brain scheduled for the first quarter of 2027

Why This Is Even Needed

The underlying logic in these projects is the same: use sun-synchronous orbits in "dawn-dusk" mode. On such a trajectory, the spacecraft moves along the boundary between light and shadow and remains almost never in Earth's shadow. This allows solar panels to operate more than 95% of the time, whereas terrestrial solar generation has a much lower average capacity factor—around 24%. For data centers where round-the-clock load is critical, such stability is no less important than the energy generation itself.

"In space there is no night, no clouds, no atmospheric losses."

This is precisely why the discussion of orbital computing is increasingly shifting from exotica to energy. On Earth, solar infrastructure needs excess capacity and storage to sustain continuous server operation. In orbit, the problem partially shifts: fewer generation pauses, less dependence on weather, and easier to predict the load profile. In the logic of project supporters, computing becomes yet another space service alongside communications, observation, and data transmission.

Where the Bottlenecks Are

The main counterargument is cooling. In space, there's no familiar air convection, which means heat from chips must be delivered to radiators differently. But supporters of the idea believe that radiator size itself doesn't look like an insurmountable problem: to radiate a kilowatt of heat into deep space, you might need less area than for a solar panel that generates the same kilowatt.

An additional signal in favor of this scheme is Nvidia's announcement of Space-1 Vera Rubin for orbital data centers and already-completed tests of the H100 in orbit. However, the economics of such an approach are still unclear. Launching a computing satellite, servicing it, and maintaining the entire system at a scale of tens of thousands of devices is far more complex than building another ground-based data center.

An even more ambitious scenario—a lunar industrial base with an electromagnetic catapult for mass satellite deployment—still looks dependent on total robotization. Without autonomous factories and robotic servicing, such a scheme would likely not be profitable even for the industry's largest players.

What Can Change the Economics

Even if orbital data centers truly appear, their fate will depend not only on rockets and solar panels, but also on the computing architectures themselves. The article mentions two directions that could dramatically reduce AI energy consumption. The first is optical computing: lab prototypes like LightGen and Taichi already show multiple gains in efficiency on specialized tasks.

The second is narrower models, including neuro-symbolic systems for robotics, which can solve practical tasks more accurately and more energy-efficiently than general-purpose neural networks. These approaches don't rule out the idea of moving part of the computation to space. Rather the opposite: the faster demand for intelligence grows, the more valuable any watt savings and every stable energy source becomes for future infrastructure.

Therefore, the debate is not about whether new chips will replace orbital servers, but about which combination of power generation, hardware, and AI architectures will prove more profitable first in practice and in the coming years.

What This Means

Orbital data centers remain a risky and very expensive bet for now, but the overall trajectory is already clear: AI is running up against energy constraints, and major companies are beginning to look for it beyond Earth. If even part of these plans works out, the next major competition in infrastructure will unfold not only between clouds, but between orbits.