Orbital Computing
Orbital Computing refers to the idea of operating large computing facilities not on Earth but in satellites orbiting the planet. There, sunlight provides power around the clock, but cooling, repairs, and data transmission are difficult.
Large computing facilities that carry out calculations for many users around the clock are called data centers. So far, they have almost always been located on Earth, in halls full of computers. Orbital Computing refers to the plan to place such facilities in satellites instead. A satellite is a device that orbits the Earth at an altitude of a few hundred kilometers. Up there, the sun shines almost continuously, and electricity from sunlight is cheap to obtain. Several companies and space agencies are therefore examining whether computing work can be outsourced into space.
AI’s hunger for power as the trigger
The reason for this idea is the enormous energy demand of modern AI systems. A single large data center can draw as much electricity as a small town. In regions such as Ireland or the US state of Virginia, power grids are therefore reaching their limits. New facilities there sometimes wait years for a grid connection. Anyone wanting new computing capacity is looking for a way out.
In orbit, two problems that are costly on Earth disappear. Solar cells there deliver significantly more energy than on the ground, because there are no clouds and no night getting in the way. In addition, the facility uses neither land nor the drinking water that many data centers use for cooling. There are also no residents complaining about noise or water consumption at an altitude of 500 kilometers.
Still, one should view the idea soberly. So far, only small test satellites exist, with the computing power of a few server racks. It will likely take many years before a facility in space can keep up with a real data center. Some experts consider the concept an expensive detour that mainly generates attention.
Solar power, radio links, and the problem of heat
An orbital data center consists of several satellites flying closely together. Each carries large solar panels and, beneath them, computer chips similar to those found in facilities on Earth. The satellites exchange data via laser beams, which achieve very high transmission rates. The results reach Earth via radio or likewise via laser to ground stations.
The most difficult technical problem is heat. Chips generate a lot of it, and in space there is no air to carry it away. Cooling there only works through radiation, meaning large black surfaces that emit heat as radiation. Such radiators are heavy and bulky, and every kilogram has to be brought up by rocket.
On top of that, there is radiation in space, which can damage chips or corrupt individual computing steps. No one can simply replace a defective component the way a technician would in a server hall. Developers therefore rely on many redundant components that can substitute for one another. Moreover, a satellite burns up in the atmosphere after a few years, while servers on Earth run for longer.
First test satellites and what makes the news
In everyday life, no one uses Orbital Computing directly yet. However, the term regularly appears in business news when tech companies present their plans for additional computing power. Google has announced a project with solar satellites, and several start-ups as well as Chinese providers have already launched test satellites. Elon Musk has also repeatedly stated that future Starlink satellites could take on computing tasks.
The technology seems most sensible first where data is generated in space anyway. Earth observation satellites capture huge amounts of images, most of which are uninteresting. If a computer evaluates the images directly on board, only the result needs to be radioed to Earth. This saves radio time, which is scarce and expensive. Experts call this edge computing, meaning computing directly at the place where the data is generated.
A common misconception is that a chatbot could one day answer from orbit. The distance is too great for that, since every signal needs time for the round trip. More realistic are tasks without time pressure, such as training models over weeks. So if you read about data centers in space, it is usually about this kind of background work.