M&A: xAI, A SpaceX Company

[thespacecow]

https://twitter.com/elonmusk/status/1984249048107508061

Simply scaling up Starlink V3 satellites, which have high speed laser links would work.

SpaceX will be doing this.

This is a reply to Eric Berger's X posts: "There has been a lot of debate about the viability of data centers in space. There are a lot hurdles, but they do seem more plausible if autonomous construction is a thing."

===

Eric Berger then wrote an article about this: Elon Musk on data centers in orbit: “SpaceX will be doing this”

===

Eric Berger then posted the link to the article on X with comment: "This seems like a pretty big deal.", to which Elon replied: "It is": https://twitter.com/elonmusk/status/1984309877217640501

===

I think this is pretty unambiguous: SpaceX will be building data centers in space, based on Starlink technololgy.

Worth pointing out that Elon oversaw the construction of xAI's data center in Memphis, which was completed in record time. He's probably the only space industry CEO who has direct hands on experience with data center construction and operations.

[RedLineTrain]

My guess is that Bezos has a good deal of knowledge, albeit not as much with AI data centers which have much higher power densities.

[steveleach]

I don't think we should be thinking about vast rooms filled with racks of computers here, but hordes of independent satellites with fast laser links between them. The Starlink model for servers.

Whether they are Starlink v3 size, launched 60 at a time, or large enough that requires a dedicated Starship launch I'm not sure about. I just don't see the need for orbital construction for something like this.

[DanClemmensen]

I don't think we should be thinking about vast rooms filled with racks of computers here, but hordes of independent satellites with fast laser links between them. The Starlink model for servers.

Whether they are Starlink v3 size, launched 60 at a time, or large enough that requires a dedicated Starship launch I'm not sure about. I just don't see the need for orbital construction for something like this.

V3-double, launched 30 at a time.

[launchwatcher]

I don't think we should be thinking about vast rooms filled with racks of computers here, but hordes of independent satellites with fast laser links between them. The Starlink model for servers.

It really depends on the workloads.   Model training performance is reportedly very sensitive to internal network latency; I would think it would be better served by a single large structure rather than a swarm of smaller satellites, unless the swarm was orbiting in very close formation.

[meekGee]

I posted elsewhere that I think this is not attractive financially (though obviously physically doable)

Curious to see how this unfolds.

[MikeAtkinson]

https://twitter.com/elonmusk/status/1984850960376791378

"Just finished a long AI5 design review with the Tesla California and Texas chip engineers. It’s going to be great.

And AI6 and AI7 will follow in fast succession.

AI8 will be out of this world."


AI5 - inference chip for cars and optimus, up to 40x faster than AI4

AI6 - hints from Elon this is likely more aimed at data centers with training features added

AI7 - - my guess a low power version

AI8 - for satellite data centres, literally "Out of this world"

Also a 6 7 joke!

[thespacecow]

Elon replied "Yes" and 🎯 to the follow X posts:

https://x.com/ApoStructura/status/1984311834631901506

SpaceX will be clustering scaled up V3 Starlinks together using laser links to form a decentralized space data center!

That will be much easier than physically assembling data centers in space, and they already have the tech.

This is huge.

https://x.com/ApoStructura/status/1984411655212802222

Data centers and cloud compute are a $1T+ market that is growing extremely rapidly.

The largest bottleneck to scaling up are permitting and energy use (which is also limited by permitting).

There is no permitting in space, there is also unlimited and constant solar energy. So space based data center make could make a lot of sense.

One of the biggest challenge to space based data center would be assembly. SpaceX can solve that by networking Starlink satellites using lasers, something they are already doing.

So instead of building one big data center they will network thousands of satellites into a decentralized one.

SpaceX has also shown that they have the ability to rapidly scale up satellite production, and to launch them for cheap.

So they are essentially extremely well positioned to become not just the leader in space based compute, but in compute overall.

This alone has the potential to be a $100B+ a year business, now combine that with regular Starlink and with Direct to Cell, and SoaceX might become one of the most profitable companies ever.

Mars is happening.

[Vultur]

Interesting. So it does sound like it's not some gigantic monolithic data center

[thespacecow]

Interesting. So it does sound like it's not some gigantic monolithic data center

As @Robotbeat pointed out on X, distributed is fine for inference, you only need monolithic data center for training.

Current estimate is 80-90% of the AI compute is used for inference: https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

As conversations with experts and AI companies made clear, inference, not training, represents an increasing majority of AI’s energy demands and will continue to do so in the near future. It’s now estimated that 80–90% of computing power for AI is used for inference.

[meekGee]

"There is no permitting in space, there is also unlimited and constant solar energy. So space based data center make could make a lot of sense."

I've never tried to permit a PV installation, but the 1 $/watt figure for large fields includes this., amd for a co-sited plant, you don't even need inverters so it's even cheaper.

But I'll accept what Vulture is saying, that it's not cost but just difficulty in execution that's the real barrier. Can't build them fast enough.

A launch of 30 V3s buses, each at 100 kWatt, is 3 MWatt. Launch 30 times, and it's a 100 MWatt installation. SpaceX can do that in a month.

But, for continuous power, you need to go to a high orbit, or use a propellant hungry precessing "terminator" orbit.  We haven't heard much about this.  Starship can't deploy to a high orbit.

I have to admit though that if SpaceX/Musk is leaning into it, I'm listening a lot more closely.

[DanClemmensen]

But, for continuous power, you need to go to a high orbit, or use a propellant hungry precessing "terminator" orbit.  We haven't heard much about this.  Starship can't deploy to a high orbit.

Of course it can. You just need to pay for a refilling Tanker flight and you can then take a maximum-mass payload to a high orbit. That extre tanker flight is cheap in comparison to the cost of this payload.

[rklaehn]

But, for continuous power, you need to go to a high orbit, or use a propellant hungry precessing "terminator" orbit.  We haven't heard much about this.  Starship can't deploy to a high orbit.

A sun synchronous terminator orbit does not require much propellant for maintenance. The precession is due to the earth oblateness. Any minor orbit adjustments can be done with electric propulsion very cheaply.

You need more delta-v to get there - probably want to refill at a low polar orbit depot - and you need launch sites that are capable of close to polar orbits, but other than that they are really good choices for many things including orbital data centers and large space stations.

[RedLineTrain]

A bit about the timing that Musk is thinking.  I've been very skeptical about the near-term advantages of orbital versus terrestrial and Starship rate limiting orbital.  But maybe Musk believes that terrestrial will grow so much as to be problematic in the medium-term.

Ryan McEntush @rmcentush 21h
each starlink v3 appears to have ~100 kW of solar (based on array dimensions). starship should carry ~50–100 per flight — call it ~6 MW nameplate per launch at 60 sats.

spacex applied for ~30k sats → ~3 GW of orbital solar in total

if starship hits even conservative projected launch rates, they’d be able to launch a nuclear reactor worth of power capacity per year by the end of the decade. unprecedented scale.

Starship could deliver 100GW/year to high Earth orbit within 4 to 5 years if we can solve the other parts of the equation.

100TW/year is possible from a lunar base producing solar-powered AI satellites locally and accelerating them to escape velocity with a mass driver.

https://x.com/elonmusk/status/1984868748378157312

[RedLineTrain]

This seems like an extreme launch rate in ~2030.  Way more ambition than previously understood.

Would test community acceptance and probably need to launch offshore.  And if launched offshore, then better larger than 9-meter diameter.

[meekGee]

But, for continuous power, you need to go to a high orbit, or use a propellant hungry precessing "terminator" orbit.  We haven't heard much about this.  Starship can't deploy to a high orbit.

Of course it can. You just need to pay for a refilling Tanker flight and you can then take a maximum-mass payload to a high orbit. That extre tanker flight is cheap in comparison to the cost of this payload.

Yes of course with refueling it can go anywhere, just double the launch cost if only one refueling can do the trick.  Triple if you need more, etc.

Musk had insight to internal launch costs and the timeline for achieving them.

If it's really $1M per launch in the 3-4 year timeframe then many things change.

[meekGee]

But, for continuous power, you need to go to a high orbit, or use a propellant hungry precessing "terminator" orbit.  We haven't heard much about this.  Starship can't deploy to a high orbit.

A sun synchronous terminator orbit does not require much propellant for maintenance. The precession is due to the earth oblateness. Any minor orbit adjustments can be done with electric propulsion very cheaply.

You need more delta-v to get there - probably want to refill at a low polar orbit depot - and you need launch sites that are capable of close to polar orbits, but other than that they are really good choices for many things including orbital data centers and large space stations.

That's what I'm reading too.

Which means instead of refueling and going to substantially higher orbits, SpaceX can put them in rings at that plane ("DD-SSO"). Plenty of room.

Amd constant sun angle also means constant radiator angle.

[Vultur]

A sun synchronous orbit also keeps them in the daytime sky from Earth, removing astronomy concerns, right?

[thespacecow]

This seems like an extreme launch rate in ~2030.  Way more ambition than previously understood.

Would test community acceptance and probably need to launch offshore.  And if launched offshore, then better larger than 9-meter diameter.

He seems to be assuming a launch cadence needed for Mars, i.e. 10 launches per day, 3650 launches per year, if 6MW per launch then that's 22GW/year. Starship V4 or later could double this by more payload capacity to LEO, and optimizing satellite design gets the rest of the way to 100GW/year.

[meekGee]

A sun synchronous orbit also keeps them in the daytime sky from Earth, removing astronomy concerns, right?

Yup, and I'm realizing the radiators not only see no sun, but also no sun-lit earth.