Google's Project Suncatcher puts AI chips into orbit to test data centres in space

Started by Gareth_11, Today at 02:40 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: Google's Project Suncatcher puts AI chips into orbit to test data centres in space   Views(Read 74 times)
Active members in this topic:
Gareth_11(1)

Gareth_11

Google has launched the first prototype satellite for Project Suncatcher, its research effort into whether AI data centres could one day run in space. The fridge sized satellite was built with Planet Labs and launched on a SpaceX rocket into a sun synchronous orbit, where its solar panels stay in sunlight almost all the time. On board are four of Google's Tensor Processing Units, the custom chips it uses for AI. KPBS has a good write up with comments from the project lead and an outside expert

The test is fairly simple for now. The TPUs will run Google's Gemma AI model in 15 minute sessions to see how the hardware copes with the stress of launch, radiation and the extreme temperature swings of space. Google plans to launch two more satellites next year to test how satellites talk to each other. The long term vision is clusters of satellites linked by lasers, each carrying dozens of TPUs, acting together as a data centre

The pitch is that space offers almost unlimited solar power and no fights with local councils over land, water or noise. Given the backlash against data centres on the ground, you can see the appeal. Google's capital spending this year is expected to reach between $180 billion and $190 billion, more than six times its 2022 level, and finding enough power is one of its biggest challenges. Space is one way to get around the grid

The project's senior director Travis Beals was refreshingly honest about the economics. He said he does not see it being cheaper in the next five years, and expects space and ground data centres to coexist for decades. Heat is a big problem, because there is no air to carry it away, so radiators are among the heaviest parts. Carnegie Mellon researcher Brandon Lucia pointed out that maintenance costs and complexity could be amplified by a factor of 10, maybe 100. Ground based data centres face none of those problems

I love the ambition, and it fits with the underwater data centre story we covered earlier. But it is hard to see this making financial sense for a long time. Would you see space data centres as visionary or a gimmick? And what happens to all those satellites at the end of their lives?