I genuinely don’t understand why anyone would put money behind this. I thought it insane when Musk first mentioned it. Can someone explain a single advantage of a space based data center? Even if it’s not worth spending this kind of money on.
google owns a fair chunk of the company from a substantial investment a decade ago. this is just a further ‘investment’ to boost spacex financials prior to the anticipated ipo. has nothing to do with elon’s pipe dream of data centers in space.
It’s going to last until some piece of space junk whacks into it, causing it to shatter into 10’s of thousands of pieces. I forsee a time when all of Musks low-orbit agenda is literally smashed apart. One errant bolt at 30k mph at a time…
Doesn’t even need that, the physics of it just doesn’t work. There’s basically no way to cool it and no way to power it. Anyone who thinks a space data center is a good idea has failed high school level physics.
It will actually not last that long, because data centers require constant physical maintenance. Its one of the many many reasons that data centers in orbit makes no real sense
They are designed to be maintenance free just like starlink. They’ll go up, and come back down every 5 or so years. They’ll constantly be iterating on them so new more powerful ones with any issues that existed resolved for the next batch.
This isn’t space based, this is traditional data center capacity that spacex acquired when it bought xAI. Although given the inclusion of a no-penalty termination clause if the required capacity isn’t provided a cynic might suspect that Elon is just juicing expected revenues ahead of the IPO and that these contracts will quietly be discarded when the promised compute doesn’t materialise…
I’d be wary of that given the detachment from reality of the TSLA share price. The market can stay irrational longer than you can stay solvent and all that.
Sadge. Probably for the best though, if Tesla is any indication, this one will stay overvalued for no good reason, long enough to put off quick shorts anyways.
I wouldn’t be surprised if he cancels all the contracts the first day possible after the IPO. He does have compute though, xAI built 2 huge data centres
Sure, the question is does he have $2.17b/month worth of spare compute to cover this + the Anthropic deal after whatever compute grok is using is accounted for
there is high military value in space based compute.
Centralized datacenters on earth are “easy targets”.
It a lot harder of low end opponents to strike a satelite, even more so if its many.
Also it makes the latency shorter to the drones on the front lines.
Imagen FPV drones, but they transmit picture strait to low earth orbit datacenter where compute happens.
This keeps the ping low and the compute hardware on the drone itself cheap.
It a lot harder of low end opponents to strike a satelite, even more so if its many.
Space datacentres are a boondoggle.
Heard of Kessler syndrome? You’ll only need to blow up one or two (or just put up a shrapnel bomb nearby) to deny LEO to everyone and take out all the current satellites. Shit, it’s already close to that point with the current satellite population (looking at you starlink). Dealing with the result will be a long and painful process with which we have little know-how and less experience.
Lasers are still light. Light travels at similar speeds through fiber as it does in a vacuum.
Except you have the added 340 miles of distance to get to their orbital height, and then the same back down. So that’s a 680 mile minimum distance for every packet, not including any transmission across the constellation which would be a greater distance since the circumference of their orbit is obviously larger than the Earth and thus landline fiber. Not to mention that every connection adds latency as well.
Is that total combination faster than a fiber landline across a Tier 1 backbone and subsequent connections to the end points? Depends a lot on the latency added by those connections and what the ground stations on either end do converting it back. It could be faster, but that’s not a definite thing.
And all of this ignores the fact that datacenters expend a shit ton of heat, which is not easy to do in the vacuum of space. The ISS which is also in LEO, ranges in temperature between -157C to +121C. So you have to take that into account when trying to dissipate heat. And the radiators on the ISS are capable of 3m²/kW. Given the power usage and heat produced by a data center… Dissipating that heat into orbit would require a ridiculously large system. Much larger than the data center itself.
It’s a ridiculous idea from every perspective due to basic physics.
If youre going around the world, you’re going through multiple connection points which will absolutely add up. Going from LA to NYC, it probably wouldn’t be.
And these data centers satellites are small. Stop thinking massive things, they’re going to be like the v3 starlink dishes but a little bigger. They’re going to launch multiple data center dishes per launch like how starlink v3 will be 60, not 1 or 2. SpaceX already knows how to radiate heat away on the satellites they launch, this isnt going to be an issue for them, they already do it today.
Edit: like size wise you need to be thinking like a server rack, plus the huge folding solar array/radiator for it. They already know how to make the folding solar panels as well. And the rack will probably be long and narrow so they can angle it best against the sun so it gets as least heat as possible from it, not a chubby thing with a lot of surface on all sides.
Edit: also land based routing isn’t direct increasing the distance likely substantially, potentially more than the up and down, the massive starlink constellation it will be more direct.
You dont need to worry about land, power, red tape or NIMBY s in space which are big problems. The sun is always shining in a sun synchronous orbit, and space is plentiful. People think these are going to be massive megawatt things as well, but they arent, they’re going to be small satellites, so each one doesnt have as much heat to deal with as people think, and they know how to radiate heat away, they already do it on their satellites.
SpaceX also knows how to make relatively cheap satellites as demonstrated with Starlink where its a mass manufacturing line for them which brings costs down.
With starlink though people actually need the service and its worth paying for if you’re in an area that needs it. Its hard to say the same about all this AI stuff, so even if they can do it, we still haven’t seen that its profitable, that enough people want it, and that people wont eventually turn to a local AI if they do want it. So its possible, but its not solving a need as big as starlink is, so it could be an utter failure even if technically possible.
Personally I dont think its going to work because they wont get enough use to be profitable, not because the idea wont work or wouldn’t work if enough people wanted it.
Edit: also SpaceX has a proven track record of being able to launch satellites at a high cadence, so its just a matter of will starship be rapidly reusable. Starship has to be rapidly reusable for this work. If it turns out its only reusable after heavy refurbishment, that really messes with the economics of everything.
I genuinely don’t understand why anyone would put money behind this. I thought it insane when Musk first mentioned it. Can someone explain a single advantage of a space based data center? Even if it’s not worth spending this kind of money on.
google owns a fair chunk of the company from a substantial investment a decade ago. this is just a further ‘investment’ to boost spacex financials prior to the anticipated ipo. has nothing to do with elon’s pipe dream of data centers in space.
THAT makes sense!
Market manipulation
An angry mob cannot destroy it during the social revolution.
However, the ground stations can still be destroyed.
It’s going to last until some piece of space junk whacks into it, causing it to shatter into 10’s of thousands of pieces. I forsee a time when all of Musks low-orbit agenda is literally smashed apart. One errant bolt at 30k mph at a time…
Doesn’t even need that, the physics of it just doesn’t work. There’s basically no way to cool it and no way to power it. Anyone who thinks a space data center is a good idea has failed high school level physics.
It will actually not last that long, because data centers require constant physical maintenance. Its one of the many many reasons that data centers in orbit makes no real sense
They are designed to be maintenance free just like starlink. They’ll go up, and come back down every 5 or so years. They’ll constantly be iterating on them so new more powerful ones with any issues that existed resolved for the next batch.
No one is repairing starlinks in space.
So it’s a boondoggle.
Wouldn’t be the first, won’t be the last.
Okay, so right, they still won’t last that long 👍 We’re saying the same thing and making the same points.
One decent hacker getting into the control and stabilization system, though…
This isn’t space based, this is traditional data center capacity that spacex acquired when it bought xAI. Although given the inclusion of a no-penalty termination clause if the required capacity isn’t provided a cynic might suspect that Elon is just juicing expected revenues ahead of the IPO and that these contracts will quietly be discarded when the promised compute doesn’t materialise…
given his reputation, previous actions, and thirst for even more wealth… anyone with more than three brain cells should see the scam being set up.
I should really buy some puts on those IPOs… They are going to plummet so hard.
I’d be wary of that given the detachment from reality of the TSLA share price. The market can stay irrational longer than you can stay solvent and all that.
Can you trade options on a stock pre-IPO? I’d imagine those would be expensive af, but I’ve been out of that game for some time
Unfortunately not, I think. At least not for us mere mortals.
I’d have to wait until it’s already being traded.
Sadge. Probably for the best though, if Tesla is any indication, this one will stay overvalued for no good reason, long enough to put off quick shorts anyways.
I wouldn’t be surprised if he cancels all the contracts the first day possible after the IPO. He does have compute though, xAI built 2 huge data centres
Sure, the question is does he have $2.17b/month worth of spare compute to cover this + the Anthropic deal after whatever compute grok is using is accounted for
there is high military value in space based compute. Centralized datacenters on earth are “easy targets”. It a lot harder of low end opponents to strike a satelite, even more so if its many.
Also it makes the latency shorter to the drones on the front lines.
Imagen FPV drones, but they transmit picture strait to low earth orbit datacenter where compute happens. This keeps the ping low and the compute hardware on the drone itself cheap.
Space datacentres are a boondoggle.
Heard of Kessler syndrome? You’ll only need to blow up one or two (or just put up a shrapnel bomb nearby) to deny LEO to everyone and take out all the current satellites. Shit, it’s already close to that point with the current satellite population (looking at you starlink). Dealing with the result will be a long and painful process with which we have little know-how and less experience.
Latency around the world is faster in space too.
If the US military is doing something across the globe, satellite laser links will be faster than any ground based system.
Lasers are still light. Light travels at similar speeds through fiber as it does in a vacuum.
Except you have the added 340 miles of distance to get to their orbital height, and then the same back down. So that’s a 680 mile minimum distance for every packet, not including any transmission across the constellation which would be a greater distance since the circumference of their orbit is obviously larger than the Earth and thus landline fiber. Not to mention that every connection adds latency as well.
Is that total combination faster than a fiber landline across a Tier 1 backbone and subsequent connections to the end points? Depends a lot on the latency added by those connections and what the ground stations on either end do converting it back. It could be faster, but that’s not a definite thing.
And all of this ignores the fact that datacenters expend a shit ton of heat, which is not easy to do in the vacuum of space. The ISS which is also in LEO, ranges in temperature between -157C to +121C. So you have to take that into account when trying to dissipate heat. And the radiators on the ISS are capable of 3m²/kW. Given the power usage and heat produced by a data center… Dissipating that heat into orbit would require a ridiculously large system. Much larger than the data center itself.
It’s a ridiculous idea from every perspective due to basic physics.
If youre going around the world, you’re going through multiple connection points which will absolutely add up. Going from LA to NYC, it probably wouldn’t be.
And these data centers satellites are small. Stop thinking massive things, they’re going to be like the v3 starlink dishes but a little bigger. They’re going to launch multiple data center dishes per launch like how starlink v3 will be 60, not 1 or 2. SpaceX already knows how to radiate heat away on the satellites they launch, this isnt going to be an issue for them, they already do it today.
Edit: like size wise you need to be thinking like a server rack, plus the huge folding solar array/radiator for it. They already know how to make the folding solar panels as well. And the rack will probably be long and narrow so they can angle it best against the sun so it gets as least heat as possible from it, not a chubby thing with a lot of surface on all sides.
Edit: also land based routing isn’t direct increasing the distance likely substantially, potentially more than the up and down, the massive starlink constellation it will be more direct.
You dont need to worry about land, power, red tape or NIMBY s in space which are big problems. The sun is always shining in a sun synchronous orbit, and space is plentiful. People think these are going to be massive megawatt things as well, but they arent, they’re going to be small satellites, so each one doesnt have as much heat to deal with as people think, and they know how to radiate heat away, they already do it on their satellites.
SpaceX also knows how to make relatively cheap satellites as demonstrated with Starlink where its a mass manufacturing line for them which brings costs down.
With starlink though people actually need the service and its worth paying for if you’re in an area that needs it. Its hard to say the same about all this AI stuff, so even if they can do it, we still haven’t seen that its profitable, that enough people want it, and that people wont eventually turn to a local AI if they do want it. So its possible, but its not solving a need as big as starlink is, so it could be an utter failure even if technically possible.
Personally I dont think its going to work because they wont get enough use to be profitable, not because the idea wont work or wouldn’t work if enough people wanted it.
Edit: also SpaceX has a proven track record of being able to launch satellites at a high cadence, so its just a matter of will starship be rapidly reusable. Starship has to be rapidly reusable for this work. If it turns out its only reusable after heavy refurbishment, that really messes with the economics of everything.