zlacker

>>g-mork+(OP)
> In the long term, space-based AI is obviously the only way to scale.

I never questioned it.

Space is also extremely cold, and if it's as dense as Musk cooling won't be an issue.

>>jopsen+X4
I can't tell how many layers of sarcasm are here, but I just want to highlight that aktshually cooling in space is quite difficult because there is no convection, so the only cooling option is radiative. Which gets a bit hard when the satellite gets blasted by the sun.

The ISS doesn't have problems staying warm, it has problems cooling off.

>>AceJoh+66
> the only cooling option is radiative.

It does say he's planning an AI sun, I'm guessing that's the temperature you need to run at for radiation to work.

>>jopsen+w8
There are already large communication satellites that consume several kW of power.

>>pantal+q9
Oh, good. So we only need to multiply that by 200 million times, per space datacenter.

>>FireBe+wb
The data center would still consist of many individual satellites, much like a earth based data center consists of many individual servers

>>pantal+ic
tbh you could just combine them with starlink sats. didn't they just apply for (and get?) a license for 1 million sats? Stick a single racks worth of gpu power on those and hey presto you've just got yourself the largest ai cluster in the world by far.

>>esskay+Zo
The problem is, they need so much more power than a Starlink satellite that they need to be placed in a sun-synchronous orbit. The orbits that Starlink satellites need to use are nothing like the orbits that would have to be used for this rejected attempt at an Ed Wood sci-fi tribute script.

And if you accept that the duty cycle of the AI+Starlink satellites will be less than 50%, it would be much better to build the data centers in random deserts and wastelands on Earth and use the Starlink network to talk to them.