[GMoromisato]

A lot (everything?) hinges on orbital data centers. If they don't work, SpaceX is overvalued.

SpaceX plans to launch 120 kW satellites, each weighing 1.7 tons. Let's be conservative and say it ends up being a 100 kW satellite massing 2 tons. Let's be conservative and say Starship can launch 50 tons to orbit for $20 million ($5 million more than a Falcon 9 launch).

50 kW per ton x $400K per ton = $8,000 per kW = $8 million per megawatt in launch costs.

That means a 100 megawatt orbital data center will cost $800 million to launch.

You need about 833 satellites for 100 megawatts, so let's round up to 1,000 satellites. Let's say one of these satellites cost $3 million (that's probably high, but let's go with that for now). That's about $3 billion for satellite manufacturing.

Bottom line: It will cost SpaceX $4 billion to launch a 100 megawatt data center.

Anthropic is paying SpaceX roughly $50 million per megawatt per year. SpaceX could sell access to its data center for $5 billion per year. Assuming the satellites last for 4 years, that's $20 billion in revenue from $4 billion in costs.

Please correct my math/assumptions, but this rough calculation shows that SpaceX could be right and Morningstar could be wrong.

[pu_pe]

> Please correct my math/assumptions

1) You forgot to include the actual costs of the GPUs and other equipment, you are just calculating launch costs which will always come on top of the other stuff

2) Datacenters require repairs and hands-on attention, and space provides extra challenges like radiation to deal with

3) They have not provided any evidence that they can solve the heat diffusion issue, which could kill the whole project

4) Musk's track record for outlandish tech is poor (hyperloop, fully autonomous driving, Mars trips, etc. were significantly delayed or not achieved yet), and unlike some of these other things no one else seems to think this is economically feasible

[naet]

I think starlink made intuitive sense; we already use satellites to transfer data for phone service and TV channels, a satellite can "see" a large area to service, the technology was there already, etc. Starlink provides a service you can't really get without going to space (coverage in remote areas).

The space data center doesn't make intuitive sense to me. Why put it in space? Wouldn't it be better just... on the ground? The technology doesn't feel like it's there either, and there would be significant competition from existing or new data centers that don't have all the drawbacks of orbiting the planet.

[ericd]

The US power grid is not able to keep up with demand, it's mired in red tape. There's also a populist revolt against DCs and the infra upgrade costs being dumped on ratepayers. That's the main benefit I see of their plan, from a US perspective. I think China probably will have a significantly easier time.

[zarzavat]

> The US power grid is not able to keep up with demand, it's mired in red tape

Why not build off-grid?

> There's also a populist revolt against DCs and the infra upgrade costs being dumped on ratepayers.

What about the other 94% of the Earth's landmass?

[ericd]

>Why not build off-grid?

He sort of is. I believe people aren't very happy with his gas turbines powering Colossus near Memphis. There's also currently a >5 year backlog for nat gas turbines for combined cycle plants.

Solar takes up an enormous amount of space to get to reasonable base load, because you have to massively overpanel in order to carry it through winter. You need a large multiple of the panels you'd need in space to get to reliable baseload, plus batteries. During grey winter weeks, our home array is frequently making like 1/6 what it does during summer, and even summer days have night/weather/etc compared to sun sync orbits.

Nuclear is hilariously expensive and slow to build. But I have some hope for small reactors and Commonwealth Fusion Systems/Helion.

>What about the other 94% of the Earth's landmass?

Geopolitical concerns, infrastructure/power availability, security, since these things are full of incredibly valuable hardware.

[zarzavat]

> Geopolitical concerns, infrastructure/power availability, security, since these things are full of incredibly valuable hardware.

The market decides that, not Elon. Do customers want to pay $X for compute in Iceland (or wherever), or do they want to pay $kX for compute in space?

What premium are customers willing to pay for compute in space? I would guess that the premium is actually negative because the quality of computation in space is worse than under the sweet embrace of Earth's atmosphere which provides a natural protection against radiation.

[ericd]

No premium, it's just a question of whether people can and are willing to build enough in those other places. Space is a bet that they can't. Also, I don't think bitflips are a huge deal in space for neural nets, we already build in randomness via temperature. A little more shouldn't make things noticeably worse. Higher energy particles could be a problem over time, but when I looked into it, it looked like the incidence of very damaging particles for a given area was surprisingly low.

[ncallaway]

Morningstar does think it's a possibility that Starship ends up being highly successful as a reusable platform and that orbital data centers end up being very successful, they just assign a 7% chance for that outcome.

[crowcroft]

I guess to this point the argument for the valuation is just, do you think Elon will beat the (very bad) odds?

I wouldn't take that bet, but I can see why some people would.

[GMoromisato]

Fair point. Even if the math works, SpaceX could fail at execution for any number of reasons.