The design aims to bypass terrestrial power and cooling limits
Remember that guy who was on here with a video that AI data centers in space would never in a million years happen because of the heat?
It's incredible how certain people who are flat out wrong can be just because it's trendy now to be against any sort of technological progress
Great work @SpaceX for leading the way 👏
Many users rejected SpaceX's orbital AI satellite compute plans as economically unviable, technically impossible, or a regulatory dodge, while some expressed excitement about pursuing ambitious space tech ideas anyway.
This guy
Exactly
I’m old enough to remember a few months ago when the timeline was filled with “data centers in space don’t work because physics” takes
@levelsio He's still absolutely right. What Elon Musk and the others are not saying is that they want datacenters in space because then they're out of the legal influence of countries.
And that makes the trouble worth it.
Here's a test of data center in space conviction...
By 2030 SpaceX AI will be building more compute:
Heat isn’t the blocker. That was peanut gallery crit.
To be a “data center” and used for training you need hundreds of thousands of GPUs just a few meters apart – not hundreds or thousands of kilometers.
If having them spread out is okay, then so would putting them into smaller racks distributed around the world like telecom.
If energy is the argument, you have to deal with the actual economics. Energy accounts for ~5% of the cost of a modern data center. It’s maybe 2% for a Vera Rubin rack.
Given the life expectancy of a satellite like this, you’ll never recoup the cost of putting it into space.
We’ll see SpaceX launch a bunch of satellites doing inference, but by the end of the decade the amount being done in space will be tiny and SpaceX will continue building more Colossus data centers on earth.
@georgecursor @levelsio I'm sad he left Twitter, but I still agree with @MalwareTechBlog on this point. The only reason why Space Datacenters make sense is because of the anticipation that you can circumvent the very real costs of regulation and political pushback (which is absolutely not guaranteed)
@levelsio They will probably make them happen, but it won’t be economically more efficient than datacenters on earth.
This whole story is just to pump spaceX for their IPO.
@georgecursor What does he say that's right? He says you can't put data centers in space because of heat. SpaceX proves you can?
@MarioHachemer @georgecursor @MalwareTechBlog Also energy cost is free
@levelsio I Was ridiculed because I said AI will be coding in assembly or even binary by a group of senior developers working for the Portuguese government and major companies.
@georgecursor Who cares if it's huge, you have space because you're literally in space, that's the point?
@levelsio He's exaggerating, you need huge radiators that are a dynamic system that angles depending on the time of day and the orbit.
Hard to do, yes. Possible, also yes.
Worth doing? Depends on the benefits like I said about international Law.
@georgecursor @levelsio @MalwareTechBlog Geosynchronous orbits aren’t exactly numerous, there’s roughly 1500 “slots” around the planet of which ~600 are already taken, and slots over densely populated areas are contested. That’s not something lends itself as a great argument.
@georgecursor @levelsio @MalwareTechBlog Not sure the 3 letter agencies that own a lot of those slots would like that.
@levelsio the thing is , even if its not possible why not try out and push it to its limit ?how else can you find out ?
While arguing about this I also realised yet another reason space datacenters make sense, everybody wants the benefits of them and they need them to be extra close for low lag but nobody wants them around.
So placing one in geosynchronous orbit over New York makes perfect sense.
@levelsio Starship was unveiled a long time ago, NASA lunar missions are in limbo now, mars missions are effectively scraped
xAI MFU on colossus training on earth was 10%
heating and interconnect is extremely hard.. i want this to work too, let's see
@MarioHachemer @levelsio @MalwareTechBlog I'm sure Elon Musk can elbow his way in the 1500 slots. He seems to get everything else.
@MarioHachemer @levelsio @MalwareTechBlog Also the datacenter can take over for whatever satellite that geosyncronous orbit was serving anyway. It's literally just an added antenna and a box on the power supply.
@levelsio Reusable rockets?! That's IMPOSSIBLE!!
The design aims to bypass terrestrial power and cooling limits
Remember that guy who was on here with a video that AI data centers in space would never in a million years happen because of the heat?
It's incredible how certain people who are flat out wrong can be just because it's trendy now to be against any sort of technological progress
Great work @SpaceX for leading the way 👏