①Google has recently proposed the concept of a space-based data center to address resource limitations on Earth by deploying its AI chips on solar-powered satellites; ②The project, named 'Suncatcher,' currently faces numerous challenges including communication, radiation resistance, and cost; ③Google executives noted, 'In the future, space may be the optimal location for scaling artificial intelligence computing.'
Cailian Press, November 5 (Edited by Zhou Ziyi) — Recently, Google presented the concept of a space-based data center in a research paper, aiming to solve resource constraints on Earth.
Due to the substantial energy consumption of AI data centers, Google is considering deploying its AI chips on solar-powered satellites and launching them into space.
Google hopes that this initiative will enable it to harness solar energy around the clock, utilizing a virtually limitless clean energy resource to achieve its ambitions in artificial intelligence without causing increased emissions from power plants or rising electricity costs, as seen with its terrestrial data centers due to surging power demands.
This research project is named 'Project Suncatcher.' If successfully launched, Google would establish the first space-based data center.
Travis Beals, Senior Director of Google's Paradigms of Intelligence division, wrote in a blog post, 'In the future, space may be the optimal location for scaling artificial intelligence computing.'
The company also released a preprint paper detailing its current progress in this field, which has not yet undergone academic peer review.
Space, the Optimal Location?
Google envisions equipping satellites with its Tensor Processing Units (TPUs) and solar panels to orbit Earth. These solar panels could generate power almost continuously, with an efficiency eight times greater than similar panels on Earth.
However, to turn this vision into reality, Google must overcome several significant obstacles.
One of the major challenges lies in ensuring that these satellites can communicate effectively with each other. Google noted that competing with terrestrial data centers requires data transmission channels between satellites capable of handling tens of terabits per second, and adjusting the satellite formation to a closer arrangement (reducing the distance between satellites to 'within a few kilometers') could help achieve this goal.
This distance is much closer than the current operational distances of satellites and is becoming a risk factor due to space debris caused by potential collisions.
In addition, Google must ensure that its TPU chips can withstand higher levels of radiation in space. The company has conducted radiation resistance tests on its Trillium TPUs and stated that these chips 'can continue to function without permanent failure under total ionizing doses equivalent to a five-year mission lifespan.'
Cost is another significant challenge. Currently, launching these TPUs into space would be quite expensive. However, a cost analysis conducted by the company indicates that by the mid-2030s, the energy costs for establishing and operating a data center in space, calculated per kilowatt annually, may be 'roughly equivalent' to those of a similarly sized data center on Earth.
Google stated that it plans to collaborate with Planet to launch several prototype satellites before 2027 to test its hardware in orbit.