Copyright Interesting Engineering
With the generative AI boom in full force, scientists have warned of the immense power requirements of data centers used to train and utilize these systems. Now, a team of researchers from NTU Singapore has joined the call to place data centers in space. Doing so would pave the way for sustainable computing, the claim. According to the team, space data centers would be powered by round-the-clock solar energy and would harness free cooling. Crucially, all of this is possible using existing technologies. Sending AI infrastructure to low Earth orbit The NTU Singapore team proposed sending data centers to low Earth orbit (LEO). This concept would be particularly useful for island cities like Singapore, where limited land leads to high real estate costs, making data centers even more expensive. In a paper published in the peer-reviewed journal Nature Electronics, the team described how satellites equipped with advanced processors could serve as orbital edge and cloud data centers. Though launching data centers to space would be expensive, the LEO environment would provide two key benefits. Firstly, data centers in space would harness the effects of natural radiative cooling due to the extreme cold temperatures. On Earth, expensive cooling systems that consume enormous amounts of water are required to run data centers effectively. Secondly, operating in space would offer virtually unlimited solar energy. The team claims these conditions would allow an orbital data center to operate with net-zero carbon emissions. “Space offers a true sustainable environment for computing,” study lead Professor Wen Yonggang explained in a press statement. “We must dream boldly and think unconventionally, if we want to build a better future for humanity.” “By harnessing the sun’s energy and the cold vacuum of space, orbital data centers could transform global computing,” he continued. “Our goal is to turn space into a renewable resource for humanity, expanding AI capacity without increasing carbon emissions or straining Earth’s limited land and energy resources.” Orbital data processing The NTU Singapore team proposed two different methods for deploying data centers in space. The first would utilize orbital edge data centers. These would harness imaging or sensing satellites equipped with AI accelerators to process raw data in orbit. By only transmitting the essential processed information to Earth, they would reduce data transmission volumes a hundredfold, significantly lowering energy requirements. The second method, orbital cloud data centers, would see satellite constellations fitted with servers, broadband links, solar panels, and radiative coolers. These would collectively perform advanced computing tasks from space. The scientists claim these methods are feasible using launch and satellite technologies available today. This is important, as AI-driven computing demand is increasing at a rapid, unsustainable pace. According to a Goldman Sachs report from earlier this year, AI-driven energy demand could rise by 165 percent by 2030. Earlier this year, former Google head Eric Schmidt told Congress that energy demand will go from 3 percent to 99 percent of total generation, due largely to rising AI demand. Schmidt, who is now CEO of Relativity Space, also claimed he acquired the launch startup to build data centers in space.
