Big Tech Wants To Build Data Centers In Space: Does This Make Sense?

Big Tech’s push for data centers in orbit rests on a simple promise—solar power, no cooling infrastructure, and global connectivity—but the physics of heat removal and radiation exposure make the “space is cold, so it’s easy” pitch far less straightforward than it sounds. Sun-synchronous orbits keep satellites in near-constant sunlight, enabling power generation without the same energy constraints that drive terrestrial data-center design. Add the ability to beam data across the globe and avoid land-use battles, and space-based computing starts to look like a clean engineering workaround.

The catch is that “cold space” doesn’t automatically solve thermal management. On Earth, processors are cooled largely by moving air to carry heat away; in orbit there’s no air, so convection fails. Cooling then depends on conduction into heat-spreading hardware and radiative cooling—emitting heat as infrared light. Radiative cooling works, but it typically requires large radiator surfaces to be efficient, which undermines some of the mass and cost advantages people associate with space. Another proposed approach circulates a refrigerant (such as ammonia) through the system and then radiates the heat away, but that still adds complexity and weight.

Radiation is the second major limiter. Cosmic rays and solar wind bombard satellites with energetic particles that can permanently damage microchips and also cause processing errors. Low Earth orbit helps reduce exposure compared with higher orbits, and mitigation strategies include shielding and using radiation-hardened chips. Those chips are built with thicker insulation and larger transistors to improve durability, but that design trade-off usually slows performance.

Even if hardware survives and heat can be managed, space computing only makes sense for certain workloads. Processing data already collected by satellites can be a strong fit because satellite-to-Earth links are far slower than modern terrestrial networks. Earth-based data centers can process at speeds above a terabit per second, while top satellite links are around 1 gigabit per second—roughly 1,000 times slower. That gap makes it efficient to filter, compress, or analyze data in orbit before sending results down.

By contrast, tasks that require huge amounts of data from the ground—especially AI training—don’t work well without major improvements in bandwidth. Training an AI model in orbit would demand far faster satellite-to-Earth communication than current systems provide.

Several efforts are already underway. In the US, Starcloud launched a satellite carrying an NVIDIA H100 chip, with its CEO Philip Johnston predicting that most new data centers will be built in outer space within a decade. Axiom Space is pursuing orbital data nodes aimed at government needs, including secure handling and defense applications. Google’s Project Suncatcher is the most ambitious: it plans to connect dozens of satellites using ultra-high-speed laser links, targeting up to 10 terabits per second, with prototype satellites planned for early 2027 and a broader vision by 2030—when data storage and retrieval could be as simple as “somewhere above Madagascar.”