zlacker

[parent] [thread] 26 comments
1. teacpd+(OP)[view] [source] 2026-02-03 00:27:58
maybe I am a fool, does space-based AI make no sense at all?
replies(6): >>idontw+92 >>fluori+e3 >>andsoi+D4 >>bandra+J4 >>baq+55 >>sleepy+oj
2. idontw+92[view] [source] 2026-02-03 00:42:06
>>teacpd+(OP)
Watts=heat

Heat has nowhere to go in space. Read about how much engineering went into cooling the ISS and now multiply that by billions.

replies(1): >>reveri+S6
3. fluori+e3[view] [source] 2026-02-03 00:49:58
>>teacpd+(OP)
No. Imagine if your computer was in space instead of being under your desk. Would that solve anything?

Orbit is a very inconvenient environment. It's difficult to reach so maintenance is a nightmare, it's moving all the time, there's nowhere to sink waste heat into, you have a constrained power budget, you have a constrained weight budget. The only things you want to put in orbit are things that absolutely can't go anywhere else.

4. andsoi+D4[view] [source] 2026-02-03 00:57:57
>>teacpd+(OP)
> maybe I am a fool, does space-based AI make no sense at all?

I think it does, for what it’s worth if we are to extend intelligence (as we know it) and potentially consciousness out there into the galaxy.

Because of distances and time, it is unlikely that humans will populate the galaxy with biological offspring (barring some technical breakthroughs that we have no line of sight on).

AI, on the other hand, could theoretically populate the galaxy and beyond, carrying the human intelligence and consciousness story into the future.

replies(2): >>macint+Z7 >>byteho+nd
5. bandra+J4[view] [source] 2026-02-03 00:58:21
>>teacpd+(OP)
Literally none. Space is the worst possible place to put something that overheats already on earth. There's probably some synergy in the other direction (AI piloting of satellites or whatever) but that's marginal at best.
replies(2): >>Walter+s7 >>mkull+t8
6. baq+55[view] [source] 2026-02-03 01:00:13
>>teacpd+(OP)
it absolutely doesn't unless there's a magical unobtainium cooling tech Musk got his hands on
◧◩
7. reveri+S6[view] [source] [discussion] 2026-02-03 01:12:52
>>idontw+92
A lot of people who are a little bit ignorant think it's really easy to cool things in space because space is notoriously very cold.

Physics, it turns out, is slightly more complicated than this and it turns out vacuum is an incredibly good insulator and more (much more) than offsets the temperature differential in terms of how easy it is to cool something.

◧◩
8. Walter+s7[view] [source] [discussion] 2026-02-03 01:16:26
>>bandra+J4
I've sure they've considered that in the engineering. For example, the solar panels would shade it. The space station has a cooling system in it. Musk's Starlink satellites don't seem to be overheating.
replies(2): >>bandra+K9 >>bdangu+7c
◧◩
9. macint+Z7[view] [source] [discussion] 2026-02-03 01:20:10
>>andsoi+D4
In, perhaps, a few hundred years.
◧◩
10. mkull+t8[view] [source] [discussion] 2026-02-03 01:23:42
>>bandra+J4
So the whole space based data center thing is just a gimmick?
replies(1): >>bandra+D9
◧◩◪
11. bandra+D9[view] [source] [discussion] 2026-02-03 01:30:56
>>mkull+t8
A gimmick, in a highly-financialized field? Surely not!
◧◩◪
12. bandra+K9[view] [source] [discussion] 2026-02-03 01:31:51
>>Walter+s7
The problem is not shading them from the Sun. And the starlink satellites run at about 1 kW
◧◩◪
13. bdangu+7c[view] [source] [discussion] 2026-02-03 01:48:53
>>Walter+s7
good read: https://taranis.ie/datacenters-in-space-are-a-terrible-horri...
replies(2): >>Walter+5f >>teacpd+Tj
◧◩
14. byteho+nd[view] [source] [discussion] 2026-02-03 01:56:55
>>andsoi+D4
Not sure if I feel comfortable with Mecha Hitler being our representative to the rest of the universe.
◧◩◪◨
15. Walter+5f[view] [source] [discussion] 2026-02-03 02:05:43
>>bdangu+7c
It is a good read. Thank you.
replies(1): >>Walter+qz
16. sleepy+oj[view] [source] 2026-02-03 02:35:29
>>teacpd+(OP)
Thinking a bit, ORBITAL ai makes little to no sense, nowhere to dump your heat, your gpus are going to be slag or only operate part of the time. But what if he put them on the moon? the lag time is what ~1.2s? That seems like an amount of time that a current AI query can take and still seem reasonable.

Not that I think it's anything but him allowing some investors to cash out when spacex goes public. Hell didn't he just shift 2billion from tesla to xai?

At the end of the day he will never see whatever bullshit he's peddling in the media about this sale his drug habit is going to kill him before then.

replies(1): >>reveri+MM
◧◩◪◨
17. teacpd+Tj[view] [source] [discussion] 2026-02-03 02:39:01
>>bdangu+7c
that's an amazing read, lots of concrete and convincing challenges; but otoh, technology is evolving at such a fast pace, maybe it is possible for breakthroughs that we couldn't imagine now to become reality sooner than we would have anticipated?
◧◩◪◨⬒
18. Walter+qz[view] [source] [discussion] 2026-02-03 04:55:38
>>Walter+5f
There's another way to look at it, though. If the data center satellites can be built and launched cheap enough, you can still come out ahead on performance/cost. I.e. if the space data center has 1/10 the performance of a ground one, and they can be built and launched for less than 10% of the cost, then you've got a business. And there are costs that won't be incurred - no electric bill, no cost for land, no charge for maintenance.

I wouldn't be too quick to dismiss Musk.

replies(2): >>reveri+mM >>bandra+KN
◧◩◪◨⬒⬓
19. reveri+mM[view] [source] [discussion] 2026-02-03 07:00:47
>>Walter+qz
* no electric bill: if you use solar panels to provide your own power, you also have no electric bill on Earth.

* no cost for land: land in sunny places where crops don't grow (for instance) is good for solar power and very cheap compared to building out a datacenter

* no charge for maintenance: sorry, I really don't get this one. Why don't the computers in space need any maintenance?

replies(1): >>Walter+2i4
◧◩
20. reveri+MM[view] [source] [discussion] 2026-02-03 07:02:52
>>sleepy+oj
Does the moon offer much heat dissipation potential vs. orbit? The lunar surface seems like an almost-as-harsh environment.
replies(2): >>reveri+PM >>sleepy+9J3
◧◩◪
21. reveri+PM[view] [source] [discussion] 2026-02-03 07:03:17
>>reveri+MM
Also 1.2 seconds is like ridiculously long, unacceptable latency.
replies(1): >>sleepy+5J3
◧◩◪◨⬒⬓
22. bandra+KN[view] [source] [discussion] 2026-02-03 07:10:35
>>Walter+qz
Why do your magic space computers not require maintenance?
◧◩◪◨
23. sleepy+5J3[view] [source] [discussion] 2026-02-03 23:18:54
>>reveri+PM
My contention is that for large ai query, it's not that unacceptable.
◧◩◪
24. sleepy+9J3[view] [source] [discussion] 2026-02-03 23:19:51
>>reveri+MM
There is a whole fucking moon you can embed heatsinks into.
replies(1): >>reveri+Gz4
◧◩◪◨⬒⬓⬔
25. Walter+2i4[view] [source] [discussion] 2026-02-04 03:12:04
>>reveri+mM
> Why don't the computers in space need any maintenance?

Because it would be too expensive to maintain them. Replacing them would be cheaper (I presume).

replies(1): >>reveri+Yz4
◧◩◪◨
26. reveri+Gz4[view] [source] [discussion] 2026-02-04 06:09:53
>>sleepy+9J3
That's a good point, the rock you're sitting on is basically a giant heat sink.
◧◩◪◨⬒⬓⬔⧯
27. reveri+Yz4[view] [source] [discussion] 2026-02-04 06:11:53
>>Walter+2i4
Ah! That makes sense. But surely there is a point at which this economics becomes true for terrestrial datacenters too (in fact I saw some glimpses of that 10+ years ago, demonstrations of shipping-container-sized self-contained units that you just plug in power, and replace the whole container when it gets degraded). If they're not doing that today for terrestrial datacenters, then it probably doesn't make economic sense yet to do it in space either.
[go to top]