Starlinks server consumes 5-7 kWh and one single B200 Rack is in a range of 100 kWh and more.

Elon Musks Colossus 1 datacenter (the small one) has 1500 Racks.

For a small datacenter, you already have to send up the equivilent of 26000 Starlink server (which is doable, to be fair, they have send 12000 up there so far since 2019).

V3 is only at 20 kWh and we have not seen V3 in space yet.

But one single rack normally has a high speed interconnect between itself and the other racks. Which means that you either have a very reliable very fast interconnect between satelites or that you have to make one satelite very big to house all of that.

Now because one Rack is 17x bigger and needs that much more energy, one GPU Rack Satelite solar area has to be 17x bigger from a area point of view.

You can't even send this payload up there without Starship btw.

So he has to solve:

- stable Operations of highend components, which are not build for space;

- Building Starship and getting it with payload into orbit

- Has to be able to send Starship up there for over 390 times to just being able to install one small datacenter

- IF Starship fails once for the 390 times, he has to rebuild Starship and the cost saving nose dives

- Needs to prove first that Starship can actually send payload up there high enough

- Needs proper groundsstations to send A LOT of data up there (Machine Learning data is a lot)

- Normally one rack has storage for tempdata and training runs; If you accept this, you still need to make sure somehow that your data is shared across full compute racks OR you need to send up a whole new Storage layer (if you mimic current datacenters) which increases everything I wrote above by 10%

Soooooo is it technical possible? Yes of course. No issue, its rocket science and we have enough technology right now to just do it besides Starship and verification that a GPU cluster can actually run for long enough in space etc.

But lets be honest here, how long do you think it will even take for Space-X to build a working Starship, enough Starships and the whole infrastructure to send 2-10 Starship up in space every single day?

This will not be more cost effective than a normal Datacenter on Earth for a very long time.

And even if you recupe some texas dessert, install massive amount of battery storage and solar and new very long (100km) fibre, this investment will be easier, cheaper and actually will stay were it is for 20-80 years without refresh.

If you estimate a lifetime of 5-10 years for a satelite setup like this, you have to send up thousends of Starships every year to just install more than once small datacenter AND to upgrade existing ones.

Btw. you do not want to now burn all of this hardware every day into our atmosphere. Its not 'free' garbage disposal.

1. you can't recylce which means we literaly burn metal away despite us having already resource issues

2. these metals actually do something with our atmosphere

Why do you give Elon such influence. He doesn’t have to do anything. SpaceX employs engineers. I think that’s one thing that’s keeping the discussion from being rational whenever it is brought up online. Too many knee jerk reactions to the personal feelings people have with Elon, quick to share sound bites they heard that reinforce their position. I don’t like the guy and even I see this as a huge game changer in the scope of space and technology. It’s no wonder to me that Google would invest in this along with other companies and soon, lots of retail investors.

Your points about whether or not some aspects are sound or safe are valid and worth bringing up but it doesn’t change the nature of how these technologies are and will be developed.