Comment by sebzim4500
8 hours ago
Cooling a datacenter in space isn't really any harder than cooling a starlink in space, the ratio of solar panels to radiating area will have to be about the same. There is nothing uniquely heat-producing about GPUs, ultimately almost all energy collected by a satellite's solar panels ends up as heat in the satellite.
IMO the big problem is the lack of maintainability.
> Cooling a datacenter in space isn't really any harder than cooling a starlink in space
A watt is a watt and cooling isn't any different just because some heat came from a GPU. But a GPU cluster will consume order of magnitudes more electricity, and will require a proportionally larger surface area to radiate heat compared to a starlink satellite.
Best estimate I can find is that a single starlink satellite uses ~5KW of power and has a radiator of a few square meters.
Power usage for 1000 B200's would be in the ballpark of 1000kW. That's around 1000 square meters of radiators.
Then the heat needs to be dispersed evenly across the radiators, which means a lot of heat pipes.
Cooling GPU's in space will be anything but easy and almost certainly won't be cost competitive with ground-based data centers.
Sure, but cooling a starlink in space is a lot more difficult than cooling a starlink on earth would be. And unlike starlink which absolutely must be in space in order to function, data centers work just fine on the ground.
This. There's no scenario where it's cheaper to put them in space.
It will cost less to put it in low earth orbit than it will be to purchase land for it at any reasonable location.
I think there's a lot of a room for an energy play that will ultimately obviate the enormously costly terrestrial energy supply chain.
1 reply →
You forget the "in 2 years" part.
According to Gemini, Earth datacenters cost $7m per MW at the low end (without compute) and solar panel power plants cost $0.5-1.5m per MW, giving $7.5-8.5m per MW overall.
Starlink V2 mini satellites are around 10kW and costs $1-1.5m to launch, for a cost of $100-150m per MW.
So if Gemini is right it seems a datacenter made of Starlinks costs 10-20x more and has a limited lifetime, i.e. it seems unprofitable right now.
In general it seems unlikely to be profitable until there is no more space for solar panels on Earth.
All kinds of industries have been conserving more each decade since the energy crisis of the 1970's.
With recent developments, projected use is now skyrocketing like never seen since.
Before that I thought it was calculated that if alternative energy could be sufficiently ramped up, there would be electricity too cheap to meter.
I would like to see that first.
Whoever has the attitude to successfully do "whatever it takes" to get it done would be the one I trust do it in space after that.
His bet then, is that the $1 million cost to get a Starlink V2 mini into orbit can be made cheaper by an order of magnitude or two.
But it is always going to be significantly more expensive than a terrestial data center. Best-case scenario it'll be identical to a regular data center, plus the whole "launching it into space" part. There's no getting around the fuel required to get out of the gravity well. And realistically you'll also be spending an additional fortune on things like station keeping, shielding, cooling, and communication.
I think that it's not just about the ratio. To me the difference is that Starlink sattelites are fixed-scope, miniature satellites that perform a limited range of tasks. When you talk about GPUs, though, your goal is maximizing the amount of compute you send up. Which means you need to push as many of these GPUs up there as possible, to the extent where you'd need huge megastructures with solar panels and radiators that would probably start pushing the limits of what in-space construction can do. Sure, the ratio would be the same, but what about the scale?
And you also need it to make sense not just from a maintenance standpoint, but from a financial one. In what world would launching what's equivalent to huge facilities that work perfectly fine on the ground make sense? What's the point? If we had a space elevator and nearly free space deployment, then yeah maybe, but how does this plan square with our current reality?
Oh, and don't forget about getting some good shielding for all those precise, cutting-edge processors.
Assuming you can stay out of the way of other satellites I'd guess you think about density in a different way to building on Earth. From a brief look at the ISS thermal system it would seem the biggest challenge would be getting enough coolant and pumping equipment in orbit for a significant wattage of compute.
Why would you need to fit the GPUs all in one structure?
You can have a swarm of small, disposable satellites with laser links between them.
Because the latencies required for modern AI training are extremely restrictive. A light-nanosecond is famously a foot, and the critical distances have to be kept in that range.
And a single cluster today would already require more solar & cooling capacity than all starlink satellites combined.
Because that brings in the whole distributed computing mess. No matter how instantaneous the actual link is, you still have to deal with the problems of which satellites can see one another, how many simultaneous links can exist per satellite, the max throughput, the need for better error correction and all sorts of other things that will drastically slow the system down in the best case. Unlike something like Starlink, with GPUs you have to be ready that everyone may need to talk to everyone else at the same time while maintaining insane throughput. If you want to send GPUs up one by one, get ready to also equip each satellite with a fixed mass of everything required to transmit and receive so much data, redundant structural/power/compute mass, individual shielding and much more. All the wasted mass you have to launch with individual satellites makes the already nonsensical pricing even worse. It just makes no sense when you can build a warehouse on the ground, fill it with shoulder-to-shoulder servers that communicate in a simple, sane and well-known way and can be repaired on the spot. What's the point?
3 replies →