> What is the benefit of "moving compute to space"?
I’ll bite. It’s cheaper and quicker to permit a launch than permit, zone and interconnect a datacenter. And solar panels in space don’t need glass cladding, which makes them cheaper to make and lift.
The downside is launch cost. But there is a breakeven between these factors that seems to have most of its error bars within Starship’s target. (By my math, around $35/kg.) So if Starship works, and all indications seem to show that it will, eventually, then that puts space-based data centers at cost parity with terrestrial ones within a decade. Which was, well, unexpected when I ran the numbers.
(The surprising finding when you run the numbers is launching the chips and solar panels isn’t the limiter, it’s launching the radiators. Which opens up whole new questions about at what scale it makes sense to stop sending those up the well.)
Ah, I was unclear. I meant build in empty land far from cities where you also have room to put in enough solar panels and batteries to power the data center.
> where you also have room to put in enough solar panels and batteries to power the data center
Environmental reviews. (The further from civilization the higher the chances the Southern farting nuknuk or whatever nests in your nowhere.) And construction costs.
The capacity of a single datacenter would require thousands of launches to get the equipment into space. I don’t believe for a second that this would be easier in any way. Cooling and bandwidth are also completely unsolved for compute on a useful scale.
> capacity of a single datacenter would require thousands of launches to get the equipment into space
But that equipment starts generating compute as soon as it’s up. This dramatically increases the capital efficiency of the venture. (Though space launch is still ultimately capital intense. Lower rates go, the more attractive it becomes.)
> Cooling and bandwidth are also completely unsolved
Quite wrong. (Though I was surprised by this, too.) ISS-style radiators (14 kg/kW) require Starship’s most optimistic launch cadences to make economic. But sub 10 kg/kw, which is closer to ISS heritage than any of the newer stuff, lets $100/kg to LEO work under most circumstances. Drop it to 6 kg/kW and even Falcon 9 becomes viable for low costs of capital (<3%) and 4-year permitting and build times.
Bandwidth is a problem, but an engineering one. (And one Starlink is working on with laser backhaul.)
Simply put, you don’t. Your DC is launched into its graveyard. If a chip burns out it burns out—maybe rack design is a bit more redundant to keep failures as independent as possible.
Maybe at some point repair is a valid optimization. But it’s not necessary for an MVP, namely, one that is competitive against 3 to 5-year terrestrial delays and sub-10% costs of capital for such projects. That’s what has surprised me.
It seems like that could change the math quite a bit, since you’d presumably be losing a lot of capacity to failures. I’d assume you would have a much higher failure rate in space, and component failure is already pretty common on earth.
I’ll bite. It’s cheaper and quicker to permit a launch than permit, zone and interconnect a datacenter. And solar panels in space don’t need glass cladding, which makes them cheaper to make and lift.
The downside is launch cost. But there is a breakeven between these factors that seems to have most of its error bars within Starship’s target. (By my math, around $35/kg.) So if Starship works, and all indications seem to show that it will, eventually, then that puts space-based data centers at cost parity with terrestrial ones within a decade. Which was, well, unexpected when I ran the numbers.
(The surprising finding when you run the numbers is launching the chips and solar panels isn’t the limiter, it’s launching the radiators. Which opens up whole new questions about at what scale it makes sense to stop sending those up the well.)