Google has dreamed up a potential new way to get around resource constraints for energy-hungry AI data centers on Earth — launching its AI chips into space on solar-powered satellites. It’s a ‘moonshot’ research project Google announced today called Project Suncatcher.
If it can ever get off the ground, the project would essentially create space-based data centers. Google hopes that by doing so, it can harness solar power around-the-clock. The dream is harnessing a near-unlimited source of clean energy that might allow the company to chase its AI ambitions without the concerns its data centers on Earth have raised when it comes to driving up power plant emissions and utility bills through soaring electricity demand.
“In the future, space may be the best place to scale AI compute,” Travis Beals, a Google senior director for Paradigms of Intelligence, writes in a blog post today. The company also published a preprint paper, which has not undergone academic peer review, detailing its progress so far on this endeavor.
“In the future, space may be the best place to scale AI compute.”
There are major hurdles Google would need to overcome to turn this plan into reality, which it explains in the blog and paper. Google envisions its Tensor Processing Units (TPUs) orbiting Earth on satellites outfitted with solar panels that could generate electricity almost continuously, making them eight times more productive than similar panels on Earth, according to Google.
A major challenge will be to ensure that the satellites can communicate well with each other. Competing with data centers on land “requires links between satellites that support tens of terabits per second,” Google writes. Maneuvering constellations of satellites into tight formations can help them achieve that, perhaps flying satellites within “kilometers or less” of each other. That’s much closer than satellites operate today, and already space junk from collisions is a growing risk.
On top of that, Google has to ensure that its TPUs can withstand higher levels of radiation in space. It has tested its Trillium TPUs for radiation tolerance and says they “survive a total ionizing dose equivalent to a 5 year mission life without permanent failures.”
It would be pretty pricey to send those TPUs into space at the moment. But a cost analysis the company performed suggests that launching and running a data center in space could become “roughly comparable” to the energy costs of an equivalent data center on Earth on a per-kilowatt/year basis by the mid-2030s. Google says it’s planning a joint mission with the company Planet to launch a couple prototype satellites by 2027 in order to test its hardware in orbit.
This is a fascinating concept! Google’s innovative approach to AI data centers in space could really change the landscape of technology and energy use. It’s exciting to see how bold ideas like this could pave the way for the future.
expand our capabilities while addressing energy concerns. It’s interesting to think about how this could also reduce latency issues by positioning data centers closer to satellite networks. The implications for global connectivity are truly exciting!
Absolutely! It’s fascinating how this concept could not only alleviate energy concerns but also open up new avenues for research in space-based technologies. Imagine the possibilities for collaboration between AI and space exploration!
up new possibilities for data processing speed. Imagine the reduced latency when data centers operate in microgravity! This could truly revolutionize how we manage and utilize AI technology.
That’s a great point! The idea of space-based data centers could not only reduce latency but also open up new avenues for sustainable energy use. With solar power being more efficient in space, it might help address some of the environmental concerns associated with traditional data centers.
Absolutely! Plus, positioning data centers in space could significantly lower cooling costs, as the cold environment of space might help manage heat more efficiently. It’s fascinating to think about how this could revolutionize not just AI, but data processing as a whole!