[kilroy123]

I was thinking about this lately as well. If you break down their plan of launching a LOT of solar panels into space and strapping a GPU cluster to it. It's not _that_ crazy now.

Prices for solar panels have dropped 90% in the last 10 years.

Price per kilo to Low Earth Orbit has dropped ~50% in the last 10 years.

SpaceX's entire plan is to keep dropping both of these prices down even more with massive solar panel / starship factories.

[amluto]

Here's the thing. Land is cheap in many parts of the world. Solar panels work okay on the ground (not as well as in space). GPUs work better on the ground (MUCH easier to cool). Connectivity is cheaper on the ground (sure, Starlink is amazing, but it doesn't hold a candle to fiber). Maintenance is possible on the ground. Reuse of old equipment is possible on the ground. Batteries are heavy, and they smooth out solar power pretty well on the ground. Oh, and natural gas, which xAI/SpaceX uses in their big datacenters, is cheap on the ground.

So, given that you're bought the GPUs, you could launch them into space, but you have the alternative option of ... not launching them into space.

[kilroy123]

But now there's massive opposition to building on land.

If you just build in space no one will be able to hold up those permits.

For what it's worth I don't disagree with you.

This depends on ridiculously low costs to orbit. But there truly is unlimited power up there from the sun.

[olyjohn]

This is exactly why Starlink is successful. If we had decent regulations, and had actual good internet on the ground, why would we bother going to outer space? The truth is that it's cheaper to build an ISP in space to work around the corruption that has kept all these monopolies in power.

[amluto]

I disagree.

Regulations or not, building a terrestrial network is expensive. Once all the permits are granted, it still costs on the order of $5/foot ($25k/mile) to buy and install fiber. (Cost varies with labor costs and by installation type, obviously.) And you need backhaul, multiplexers or switches, etc.

In a fairly dense area with a lot of customers and proximity to existing transit providers, fiber is very cost effective. In a rural area with long last-mile links and few customers, fiber is more expensive and satellite can win.

Starlink has another potential issue in dense areas: limited bandwidth. The beams from satellites that are near each other can interfere with each other, so doubling the number of satellites may not double the aggregate bandwidth available in a given small, densely populated area. In contrast, fiber scales better than linearly up to huge bandwidth: installing 10 strands of fiber is much less than twice as expensive as installing five strands, and the bandwidth available per single-mode strand has been increasing over time without any requirement to use fancier fiber.

[lolc]

I'm under the assumption that Starlink will never provide more than one percent of the mobile bandwidth in a developed country. Because they can't match dense antenna webs in cities. Am I wrong?

[xboxnolifes]

Well outside my field, but I was under the impression that getting the energy into the GPU (solar) was not the problem. The problem is getting the heat out. It takes a lot of surface area to radiate out the heat that GPUs produce. Maybe it ends up economical, but just having it in space adds in a bunch of challenges, not least that once it's up there you can't send in a technician to troubleshoot.

And if solar panels drop another 90%, why not just slap them on earth?

[hackernudes]

> if solar panels drop another 90%, why not just slap them on earth?

Because you need batteries then.

[stephc_int13]

Not the same solar panels. As with most equipment, the constraints are very different in space/orbit, and the suppliers are not the same.

[olyjohn]

It's still going to be what, $100,000 per ton to launch on Starship?

The AI1 Satellite that SpaceX has shown, is basically one rack. Something like 72 GPUs. Elon himself said it was a "rack in space" more than a datacenter in space. Sure I guess you could launch 100,000 of them and call it a datacenter. But their own graphic shows this thing to be 2 to 2 1/2 tons. So who is gonna spend $200,000 to launch a single rack into space? Get the panels as cheap as you want, the launch cost is still pretty massive just to get 150kw of free power.

I mean maybe my napkin math is way wrong here, but I don't see what the cost savings is.

[orwin]

It won't work btw, unless they don't put the satellites in LEO. Earth radiate away big time (it's like 200K in earth shadow in LEO

[einpoklum]

"solar panel / starship factories"

Oh, sure, both kinds of factories are legitimately slashable as they are totally the same thing. Pretty much. Right? :-(

[qaq]

it is crazy