Support NeoGAF

[Unknown Soldier]

I really would need Tesla to be merged into SpaceX

 

[Mengy]

I saw a comment suggesting data centers in space might be a new Musk Hyperloop. A means of setting some cool sounding flag off in the far distance and using it to raise value.

Throw "in space" at the end of something and you have a cool sounding idea.

I'm sure he'll feed confidence using one of his "It's not that hard" lines knowing that most people don't have a fucking clue about the challenges of such a task. See self driving cars by 2015. Tunnels to solve traffic problems, Vacuum tube trains.

I'm certainly not saying the notion is impossible, almost everything is possible. But practicality, cost and setting a goal that feels closer than actually is, reeks of a Musk sounding sales pitch.

Cribbed this comment from a NASA person


Google published their own findings

Exploring a space-based, scalable AI infrastructure system design

research.google

SpaceX really is going to put data centers in space. It's happening, and it does address a lot of the issues ground based data centers are encountering today. However, its also not a simple nor easy goal either.

Let me address the points raised by the NASA guy in your post:

1. Getting things to space is incredibly expensive
A: That has traditionally been true, but SpaceX can put mass into orbit at a fraction of the costs for NASA or other agencies.

2. Ingress/egress are almost always a major bottleneck - how is bandwidth cheaper in space?
A: With Starlink as the intermediary bandwidth won't be an issue.

3. Chips must be "Rad-hard" - that is do more error correcting from ionizing radiation - there were entire teams at NASA dedicated to special hardware for this.
A: Yes, but radiation creates random bit errors in files, which won't be an issue for LLM AI as their large datasets are already very resilient to bit flips here on Earth. Radiation errors shouldn't pose much of an issue.

4. Gravity and atmospheric pressure actually do wonders for easy cooling. Heat is not dissipated in space like we are all used to and you must burn additional energy trying to move the heat generated away from source.
A: True. The satellites will use radiative cooling, which we've been doing on satellites for decades. However, these will be high power satellites, and while radiative cooling will work, it will be tricky to design it into a compact and efficient satellite. Not impossible, but challenging.

5. Energy production will be cheaper from earth due to mass manufacturing of necessary components in energy systems - space energy systems need novel technology where economies of scale are lost.
A: No. Solar panels are more efficient in space when they get direct sunlight without the atmosphere in the way. Like on the order of 4-5X more efficient. Power for the AI satellites will be relatively easy to design.