As artificial intelligence (AI) spreads, power consumption at data centers is rising quickly, intensifying competition to move servers and semiconductors into space. The idea is to perform AI computation in space by generating power from solar energy there, without burdening terrestrial power grids and water supplies.
On the 30th (local time), the California Institute of Technology and space computing startup Sophia Space said they had obtained a U.S. patent related to cooling and structural technology for space data centers. The researchers proposed a "tile (TILE)" structure that combines solar cells, computers, storage, communications equipment, and a radiator into a single thin module.
◇ Why cooling is difficult in cold space
One of the technical challenges for space data centers is cooling. It is easy to assume that cooling computer heat would be simple because space is cold, but the reality is the opposite.
Ground data centers remove heat by circulating cold air or water around chips. But in the vacuum of space, there is no air or water to carry heat. That is because there is no convection, where moving air transfers heat.
Therefore, heat generated inside a spacecraft must travel through metal plates, thermally conductive materials, and heat pipes to an external radiator and then be emitted as infrared radiation. The radiator converts heat into invisible light and sends it into space.
In particular, to radiate a lot of heat at low temperatures, the radiator must be very large. Operating the radiator at high temperatures can reduce its area, but it can shorten the life and degrade the performance of semiconductors and memory. Conversely, to keep chip temperatures low, a wide and heavy radiator must be transported into orbit.
◇ Solar cells on one side, radiator on the other: a "tile-type" space data center
In the "tile" structure patented by the Caltech and Sophia Space team, solar cells are placed on one side of a flat tile and a radiator on the opposite side. Electronic components such as processors and memory are distributed between the two faces.
Heat generated by chips in the tile structure moves through a heat-spreading layer directly to the radiator on the back. The patent includes using pyrolytic graphite and heat pipes as heat-spreading materials. Pyrolytic graphite spreads heat rapidly in a specific direction, and a heat pipe is a sealed device that transfers heat without a separate pump.
The researchers also presented a design that places a thermal barrier zone of empty space or insulation between the two areas to prevent heat from the solar cells from being transferred to the computer side. In this configuration, the solar-cell face of every tile points toward the sun, while the opposite radiating face is kept in a direction favorable for heat emission.
However, obtaining a patent acknowledges the originality of the technology and the scope of rights; it does not prove that the heat of high-performance AI chips can actually be handled stably over long periods in orbit. Sophia Space plans to verify the performance of a small module through an orbital demonstration in 2027 and build a larger system in the early 2030s.
◇ Research into chips that generate less heat themselves
Recently, space data center research has also been moving toward reducing heat generated by the computer itself. A team at Iowa State University in the United States proposed in June applying "compute-in-memory (CIM)" technology, which performs calculations inside or near memory without repeatedly moving data between memory and the processing unit, to space-use AI Semiconductor.
Conventional graphics processing units (GPUs) repeatedly move data between the processing unit and high-bandwidth memory. In simulations, when the processing unit and high-bandwidth memory were placed in different locations, heat concentrated at specific points. When radiative capacity was insufficient, chips automatically throttled their operating speed to lower temperature. By contrast, the CIM structure distributed heat relatively evenly, and under limited heat-dissipation conditions delivered higher compute performance per unit of power.
The researchers said this means that when designing semiconductors for space data centers, engineers must consider not only peak compute performance but also where heat occurs and how it is distributed.
References
arXiv (2026), DOI: https://doi.org/10.48550/arXiv.2606.05741