Moving The Cloud to orbit

[DreamyPickle]

I approached this with a lot of skepticism but after some reading I think there might a real business case here. It's certainly the most plausible extension of the current space economy beyond current markets of communications and observation.

Traditional data centers are largely bound by IO capability - they store and retrieve data but do relatively little actual number crunching. Therefore they benefit from placement next to internet exchanges and population areas. AI data centers are much more focused on "compute" and external communication latency is much less important. Scaling up compute results in higher power usage and we're starting to see a trend where electricity and cooling become limiting factors. This is not speculative - these trends are seen in the construction of terrestrial data centers. Here's a very interesting video on the subject:

https://www.youtube.com/watch?v=dhqoTku-HAA

The key advantage of launching data centers into space is that a Dusk-Dawn Solar-Synchronous Orbit can give you continuous solar power on a 24/7 basis without the need for storage. That power can then be consumed on the satellite itself to turn inbound queries into AI-powered results. This ability to "covert electricity into money" without transporting a physical product is new. The closest equivalent is scam mining, but the scale is much larger.

It's similar to the business for "space solar power" except there is no need actually transmit the power. It removes the hard part!

The big problem is that you are competing directly against terrestrial data centers and power generation and there are many ways to generate power cheaply if location is not very important. For example Iceland has abundant geothermal power and is currently using it to smelt aluminum - this could be replaced with running AI data centers instead.

The dominant power source now is a combination of solar and natural gas peaker plants but storage costs are dropping fast so eventually we will see a competition between "space solar panels + radiators + launch" versus "10x the amount of ground solar panels + batteries". There are many strong factors pushing the development of battery storage, it's hard to bet against it. Why launch satellites when you can cover Arizona and Nevada with solar panels and batteries instead? Amusingly you would have to use very similar closed-loop cooling technologies.

Space-based telecommunications have the advantage of reaching the entire surface of the Earth. It doesn't have to be better than all earth-based communication technology, it only need to be better at reaching in enough places that it is profitable. No such advantage exists for AI data centers - satellites competes equally with all ground-based facilities.

The very long term argument is that space-based AI consume power than never hits the Earth and can be scaled infinitely. But this is quite weak as long as all the solar power hitting the Earth's deserts is wasted.

SpaceX is obviously ideally suited to try and make this work. It would be very interesting to see how cheap they can bring the launch of AI compute rack with Starship and a Starlink-derived bus.

There are claims that AI is a bubble and when it pops all this compute will be useless. Valuations will fluctuate wildly and investment might even decrease but I believe that AI already generates genuine value. If anything the limitation is that humans don't yet know how to fully take advantage of AI capabilities.

[InterestedEngineer]

There are claims that AI is a bubble and when it pops all this compute will be useless. Valuations will fluctuate wildly and investment might even decrease but I believe that AI already generates genuine value. If anything the limitation is that humans don't yet know how to fully take advantage of AI capabilities.

This is worth a dive into.  Some numbers first

Worldwide software sales in 2024:  $700B
Gross Margin: 80%
Worldwide net margins for software sales in 2024:  15%

Now the software industry is weird, they don't attribute to COGS (Gross Margins) most of their costs, so we'll say their real "COGS" (aka programmers, data center costs, etc) is 80%. (aka 3/4 of their net costs).

This makes it 700B * 0.8 = $560B in costs - programmer salaries, compute, buildings to house them, etc.

Now I've proven and industry stats have shown that Claude Code makes a good developer more than 2x as productive, and the cost is rounding error in the above numbers.   So that's $280B worth of value.

Now one could argue that software is a very elastic demand, so you can take it as "we'll save $280b/year" or "the market will go up by $280B".  It doesn't really matter.

Let's round that to $300B for ease of calculations and acknowledgement of how fuzzy this is.

What is $300B/year worth in terms of capital expenditure?

The basic economic equation is CapitalValue = CashFlow/CapRate

Now one could argue what the CapRate is, it's somewhere between 5-10%, but the "risk adjusted tech productivity" rate is closest match and that's ~7.5%.

$300B/0.075 = $4T capitalization (range 3-5T).

So just for the software industry ALONE, the capital to throw at this is $4T.

The current market cap for AI infrastructure, is on the order of $10T.  So it's not WAY off, and I haven't included any non-software industries such as self-driving cars, the fact that the software industry is only a fraction of the actual software produced (Tesla isn't considered a software firm for example, yet produces a lot of software)

So I'm not sure there's going to be an AI crash.  If there is, it'll be like 2000-2002 - over quickly, and taking off like a rocket post crash.

I should note that at $4 capitalization what SpaceX is currently doing is rounding error.  If SpaceX seriously gets into space-based AI data centers it'll be doubling its capitalization.

[StraumliBlight]

1276-EX-CN-2025 [Dec 30]

Planet seeks experimental authorization, under Part 5 of the Commission’s rules, to launch and operate the experimental M1 NGSO satellite in Low Earth Orbit to conduct a testing and demonstration mission to explore building scalable machine learning compute systems in space. The M1 satellite is a prototype for a potential future constellation of spacecraft intended to help address the escalating demand for artificial intelligence compute by creating a scalable AI infrastructure in space. The spacecraft will incorporate a payload of purpose-built AI computing chips provided by Planet’s customer onto a Planet-designed spacecraft bus and is intended to obtain preliminary test data on how the payload hardware operates in space. This first prototype for new in-orbit infrastructure will inform the design of potential follow-on missions.

[...]

The M1 spacecraft is a smallsat with a launch mass of no greater than 220 kg. Basic physical dimensions are approximately 2895 mm x 1840 mm x 650 mm in the deployed configuration.

[...]

The M1 spacecraft will be launched into a circular SSO orbit as a secondary payload with an injection altitude of 590 km +/- 20 km

Suncatcher prototype?

[StraumliBlight]

Tech Crunch: Why the economics of orbital AI are so brutal [Feb 11]

In a first analysis, today’s terrestrial data centers remain cheaper than those in orbit. Andrew McCalip, a space engineer, has built a helpful calculator comparing the two models. His baseline results show that a 1 Gw orbital data center might cost $42.4B—almost three times its ground-bound equivalent, thanks to the up-front costs of building the satellites and launching them to orbit.

Changing that equation, experts say, will require technology development across several fields, massive capital expenditure, and a lot of work on the supply chain for space-grade components. It also depends on costs on the ground rising as resources and supply chains are strained by growing demand.

The key driver for any space business model is how much it costs to get anything up there. Musk’s SpaceX is already pushing down on the cost of getting to orbit, but analysts looking at what it will take to make orbital data centers a reality need even lower prices to close their business case. In other words, while AI data centers may seem to be a story about a new business line ahead of the SpaceX IPO, the plan depends on completing the company’s longest-running unfinished project—Starship.

https://andrewmccalip.com/space-datacenters

Useful cost calculator.

[sdsds]

Tech Crunch: Why the economics of orbital AI are so brutal [Feb 11]

In a first analysis, today’s terrestrial data centers remain cheaper than those in orbit. Andrew McCalip, a space engineer, has built a helpful calculator comparing the two models.[...]

https://andrewmccalip.com/space-datacenters

Useful cost calculator.

The website is indeed useful, not only for the calculator but for the right-hand column of "References," which appear to be mostly supporting material generated by Gemini, ChatGPT, etc. Some of those include links to further supporting academic material.

[Vultur]

Thing is, the decision to do orbital data centers or not may not be driven by an objective analysis of cost per kW of compute power/payback time/profitability vs expenses/etc. We're pretty likely in a situation of "irrational exuberance" with the AI/data center boom.

If the boom lasts long enough they'll probably get a lot of investment & be built even if they aren't actually cheaper or better. If the boom ends before that...

[InterestedEngineer]

Tech Crunch: Why the economics of orbital AI are so brutal [Feb 11]

In a first analysis, today’s terrestrial data centers remain cheaper than those in orbit. Andrew McCalip, a space engineer, has built a helpful calculator comparing the two models. His baseline results show that a 1 Gw orbital data center might cost $42.4B—almost three times its ground-bound equivalent, thanks to the up-front costs of building the satellites and launching them to orbit.

Changing that equation, experts say, will require technology development across several fields, massive capital expenditure, and a lot of work on the supply chain for space-grade components. It also depends on costs on the ground rising as resources and supply chains are strained by growing demand.

The key driver for any space business model is how much it costs to get anything up there. Musk’s SpaceX is already pushing down on the cost of getting to orbit, but analysts looking at what it will take to make orbital data centers a reality need even lower prices to close their business case. In other words, while AI data centers may seem to be a story about a new business line ahead of the SpaceX IPO, the plan depends on completing the company’s longest-running unfinished project—Starship.

https://andrewmccalip.com/space-datacenters

Useful cost calculator.

nowhere on it does it show that turbine components are back-ordered to 2030.

Permitting costs and delays

Just the row cost of installed turbine - you should be able to 10x the price, because that's the realistic cost of "install one within 90 days".

There *is* such a thing as missing a market window, and they show none of that.

In short, it's an engineering nerd's calculator, not a market-aware calculator.

[Paul451]

nowhere on it does it show [...] Permitting costs and delays

And is that not a thing for launches?

[Twark_Main]

nowhere on it does it show [...] Permitting costs and delays

And is that not a thing for launches?

"Is also a thing" and "has the same cost impact" are very different.

Yes launches also have permitting. No the costs are not as high compared to site-by-site environmental permitting.

[Paul451]

No the costs are not as high compared to site-by-site environmental permitting.

Source?

[InterestedEngineer]

No the costs are not as high compared to site-by-site environmental permitting.

Source?

The entire compendium of Nasaspaceflight forum threads.

If you didn't see the drama of the site permitting, you missed some epic stuff.

But now that the site(s) are permitted, the launch licenses are far easier to get.

[Paul451]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

[InterestedEngineer]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

cost is near irrelevant.

compare schedules.

Did you watch Elon's latest interview?  He laid it all out.   Years for permits.  4 years for turbine parts

https://www.youtube.com/watch?v=BYXbuik3dgA

[Robotbeat]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

A 100GW datacenter?

[InterestedEngineer]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

A 100GW datacenter?

at a conservative 5MW per launch, and given *current* EIS permits (the long pole for any project) of 145 Starship launches per year already approved between the Cape and BC, that's 725MW/year *already permitted*.

Given a 3 year time frame and the "foot already in the door" effect, 10xing that is easy. 

so that's > 7GW/year already in the pipeline permit.

That's the equivalent of 3-7 terrestrial data centers that will have a longer permitting process, you can't get turbine blades for till 2030, etc.

And it's the capability to put 3-7 1GW+ class terrestrial data centers into orbit *every year*. starting in 2028-ish.

Contrast that to zero such data centers in came online in  2025 and 5-8 will likely come online in 2026.

2027-2028 has 13 in the pipeline, and that's over 2 years, so the rate is flat.  Which one expects if there are severe resource constraints.

So TL;DR - Elon will be able to single-handedly double the data center growth rate of the USA in 2028.

[Lee Jay]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

A 100GW datacenter?

at a conservative 5MW per launch, and given *current* EIS permits (the long pole for any project) of 145 Starship launches per year already approved between the Cape and BC, that's 725MW/year *already permitted*.

Given a 3 year time frame and the "foot already in the door" effect, 10xing that is easy. 

so that's > 7GW/year already in the pipeline permit.

That's the equivalent of 3-7 terrestrial data centers that will have a longer permitting process, you can't get turbine blades for till 2030, etc.

And it's the capability to put 3-7 1GW+ class terrestrial data centers into orbit *every year*. starting in 2028-ish.

Contrast that to zero such data centers in came online in  2025 and 5-8 will likely come online in 2026.

2027-2028 has 13 in the pipeline, and that's over 2 years, so the rate is flat.  Which one expects if there are severe resource constraints.

So TL;DR - Elon will be able to single-handedly double the data center growth rate of the USA in 2028.

Yeah?  You're assuming Starship gets working soon, and that manufacturing all those satellites is possible.  Where are you going to get >7GW a year of space rated solar panels, for example?  That's got to be several orders of magnitude more than can currently be manufactured.

[InterestedEngineer]

Yeah?  You're assuming Starship gets working soon, and that manufacturing all those satellites is possible.  Where are you going to get >7GW a year of space rated solar panels, for example?  That's got to be several orders of magnitude more than can currently be manufactured.

What do you mean by space rated solar panels? What do you know about that tech?

[Robotbeat]

That's launch sites.

You were comparing terrestrial data centres with orbital equivalents.

Show me that the cost of and delay from launch and communication licences for a million satellite constellation, for example (presumably with the 10k+ launches per year that Musk keeps talking about), is lower and less than the one-time permitting/etc for an equivalent terrestrial data centre.

A 100GW datacenter?

at a conservative 5MW per launch, and given *current* EIS permits (the long pole for any project) of 145 Starship launches per year already approved between the Cape and BC, that's 725MW/year *already permitted*.

Given a 3 year time frame and the "foot already in the door" effect, 10xing that is easy. 

so that's > 7GW/year already in the pipeline permit.

That's the equivalent of 3-7 terrestrial data centers that will have a longer permitting process, you can't get turbine blades for till 2030, etc.

And it's the capability to put 3-7 1GW+ class terrestrial data centers into orbit *every year*. starting in 2028-ish.

Contrast that to zero such data centers in came online in  2025 and 5-8 will likely come online in 2026.

2027-2028 has 13 in the pipeline, and that's over 2 years, so the rate is flat.  Which one expects if there are severe resource constraints.

So TL;DR - Elon will be able to single-handedly double the data center growth rate of the USA in 2028.

Yeah?  You're assuming Starship gets working soon, and that manufacturing all those satellites is possible.  Where are you going to get >7GW a year of space rated solar panels, for example?  That's got to be several orders of magnitude more than can currently be manufactured.

wait until you learn about how SpaceX already makes their solar panels using regular old monosilicon cells

[Robotbeat]

Yeah?  You're assuming Starship gets working soon, and that manufacturing all those satellites is possible.  Where are you going to get >7GW a year of space rated solar panels, for example?  That's got to be several orders of magnitude more than can currently be manufactured.

What do you mean by space rated solar panels? What do you know about that tech?

Apparently nothing because SpaceX has been making their own Starlink solar panels using regular terrestrial solar cells for this entire time.

[spacenut]

Seems to me there are several questions to this whole situation.  Musk mentioned it would take about 4 years to build a data center and get the permitting and equipment for a gas generator powered data center.  When if Starship gets operational Musk mentioned the cost/kg to get to orbit will not be $500 as mentioned but probably around $200/kg or less and could begin launching immediately.  SpaceX could also launch a combination Starlink/AI satellite system to orbit.  Thus vertical integration.  Sure the core data center would be in an orbit always in sunshine, but Starlinks in other orbits could be useful in transferring data back and forth. 

The second, wouldn't it be easier to get Starships fully operational and work out all bugs in the meantime by having a lot of launch history?  As someone said, Starlink, Starthink, and a lunar facility are all going to have to have Starships anyway, so they can be launched and built during the off synod for Mars transportation.  Mars during 6 months out of about every two years would have Starship flotillas traveling to Mars.  During the other 18 months or so build up and build out Xai satellites and a lunar base of operations/manufacturing/rail launch facilities.  This would keep Starship/Superheavy launches going all the time, thus decreasing operational costs unlike SLS only once a year and having it's own overhead costs making launch costs extremely high. 

I just think the comparison is not as truthful as the report indicates.  Building an earth based facility would be best in the desert to take advantage of solar power, but water for cooling would be a problem.  East of the Mississippi a large data center is harder to permit, get power, but water for cooling might be better if located near a river.  Then people are not wanting to destroy agricultural land and they think the gas turbines are polluting even if only CO2 and 2H2O's are the waste from burning.