China Is Building The First “AI-Powered” Data Center In Space –Why?

It might seem a bit odd at first, if not a bit sci-fi, but there are some very good reasons to place data centers in orbit. There are, of course, massive hurdles and problems to overcome, too. This initial launch might end up shaping how these possible systems grow, but also how we use Large Language Models (LLM), the correct term for what is popularly referred to as “artificial Intelligence”.  

Why a data center in space?

The data centers that host these programs require a lot of energy and a lot of water for cooling, which has become a major strain on local resources on Earth. The possibility of relocating them to space is seen as both extremely intriguing and environmentally and socially friendly.

“The power consumption of data centers around the world is spiking massively with the introduction of these ‘AI’ systems and programs, which is obviously becoming a bit of concern for generation capacity, carbon emissions, etc,” Russell Hills, a spacecraft system engineer who is not involved in the array, told IFLScience.

According to the Chinese government, the satellites will use the cold vacuum of space for cooling while the supercomputer satellite array crunches data with a combined capacity of 1,000 peta (1 quintillion) operations per second. Other companies have started designing satellites where these huge amounts of raw data processing can be done in space, powered by the Sun and releasing their waste heat into space, thereby lowering their carbon emissions. 

The possibility of using the Sun (in space) to power the data centers is certainly appealing, but in this case, the array is not intended to investigate approaches to power generation. Instead, China is investigating testing how data processing can be done in orbit.

“They’re kind of quite specific about the fact that it’s not so much they’re putting data centers in space because that’s a better place for data centers to be. It’s because there is a need for data centers and AI to be in space to work on space-based data,” Hills told IFLScience.

It’s also easier than ever to send a satellite into space, and these satellites are collecting more data than ever before; data that needs to be transmitted down to Earth before it can be analyzed. The idea for this constellation is to have the data center directly in space, so all the analysis is done in orbit, and the precious transmission on limited bandwidth is just the useful data. This also has the possibility of making the whole process faster.

“Taking a simple example, if you’re using a surveillance satellite rather than constantly downloading the images of a certain region of interest, you can run image identification algorithms, change detection algorithms, etc. And you can get those images that just say ‘oh, something’s changed,’” Hills explained.

Less data to download might mean also a smaller ground station, and more flexibility in how and where the data gets to you. 

While the advantage is clear, the challenges are present, and some are very tough to solve. Doing anything in space is costly and complicated. It is a lot more difficult to design a complex machine to work in space.

Other proposals for data centers in space

The Chinese data center constellation is using a small amount of energy, but a data center processing satellite data for GPS systems, telescopes, weather forecasting, communications, etc, like the ones we use on Earth, would need a much larger amount of energy. China is not the only nation to experiment with supercomputers in space, however.

One proposal by Washington-based Starcloud sees a solar panel structure of 4 kilometers by 4 kilometers (6 square miles), which is about one-fifth of the area of Manhattan. We have never built anything close to that size. Former Google CEO Eric Schmidt even reportedly bought Relativity Space just to put data centers into space.

Europe, too, has a project to look into the possibility of data centers in space. While not planning any megaengineering in space just yet, the program might still be underestimating the cost of such an endeavor. In a previously shared white paper, their power generation costs were based on optimistic projections, including a massive reduction in the cost to access space with a not-yet-deployed Starship.

Whether the future of data centers is in space remains to be seen. China, however, is the first to carry out an empirical test on an operational scale.