back
117 comments
My draft blog post about all the things that are up with space data centres keeps getting bigger and bigger.

  "The other half of the MS model is data centres. “Orbital compute deployments” start in 2028, reach cost parity with their earthbound equivalents by 2031, and put 364GW of rigs in space by 2040."
With 25% efficient cells, at 500 km altitude, in a terminator-tracing SSO, this is enough to occupy a *contiguous* ring roughly 25 m tall, all the way around that orbit.

Also, from other statements they're clearly copying Alphabet's study which said cost parity in 2035, if they can actually launch 370,000 tons and maintain their learning rate.

https://arxiv.org/pdf/2511.19468

  "A $668bn funding obligation to 2034 that delivers free cash flow that year of negative $48bn sounds less than ideal, though FCF might flip positive to $138bn in 2035 if everything goes to plan, so that’s nice. The SpaceX CEO presumably has a long history of delivering products on time and to the required specification that can support such confidence."
I love the snark here.

  "Helium-3 is one of the clearest examples of why lunar infrastructure could matter. The isotope is extremely rare on Earth, with current supply largely tied to tritium decay, but the Moon has accumulated helium-3 for billions of years because it lacks Earth’s atmosphere and magnetic field. NASA mining concepts often assume concentrations around 20 parts per billion, meaning helium-3 is abundant in total but painfully diffuse, requiring hundreds of tons of regolith to be mined and heated to recover small quantities."
Ugh. This will need a separate blog post for why it's stupid. At 20 ppb, even if we could fuse He3, that makes lunar regolith marginally less energy dense than firewood. Also, anyone with a fusion reactor can make He3, even highschool students with home-made fusors can already do this. I'll have to check sources and maths to make sure I've not missed something important about which would be cheaper, *currently existing* neutron sources like fusors or going to the moon, but regardless, we can't currently use this stuff for fusion and the moment we can we won't need to mine it.

(I have not yet formed an opinion about non-fusion uses for He3).

I just skimmed that linked paper. Only mention I found of cooling is:

> Cooling would be achieved through a thermal sys- tem of heat pipes and radiators while operating at nominal temperatures.

Isn't that drastically underselling potentially one of the harder parts of this whole endeavor?

I view articles like that as a kind of roleplaying, essentially. The authors are pretending to be space hardware engineers, but the results are not remotely realistic.
> Isn't that drastically underselling potentially one of the harder parts of this whole endeavor?

Everything in space is hard; but these are Alphabet researchers not NASA researchers, and honestly even the NASA papers I've been skimming through have a lot of simplifying assumptions in them, so that's not something to hold against them here.

They are just saying when they think it's worth considering, after all, not giving a detailed all-aspect proposal for how to make one.

Very drastically, the ISS solar panels can generate up to 120kW of power, look at the size of its radiators needed to cool it down.

Scaling that to the hundreds of GW range is quite laughable.

https://x.com/SawyerMerritt/status/2064108916611420273?lang=...

While I'd suspect the design is still in flux, the current design is for a 120kw satellite with 110 square meters of radiators. Scaling to hundreds of gigawatts is intended to be by repeatedly launching smaller designs.

The cost of fuel is not the problem. We have unlimited uranium and thorium, and there is no reason a fusion should be cheaper. Sure its more dense then fission but fission is absurdly dense already and the economics don't improve that much.

The idea that it makes sense to use moon based He3 compared to using thorium that is already mined in waste quantities is absurd. Thorium is free energy already and the machine that turns it into energy is simpler the any fusion reactor we can come up with.

I'm not very informed on SpaceX plans, but one thing I think people gloss over is how much maintenance a data center requires. Parts fail, computers get stuck on crash loops, etc. A space data center would need workers - computer people, not astronauts - and a constant supply of parts tob replace hardware. Whoever wrote this proposal doesn't understand neither space nor data centers.
Is a data center satellite really that different from a communications satellite? Starlink sats must have some significant processing power and nontrivial control system, and they work without physical maintenance. One data center sat is like one server rack, if it fails, it's fully lost and you just deorbit it, as it's done with Starlink sats. They sent 12443 Starlink sats to space, deorbited 1684. The thing that matters is failure rate, and the economics resulting from that. And also the cost of specialized resilient hardware.
I would pay so much money just to go on a mission to replace a bad stick of RAM on a data center in space.
I'm also writing a blogpost on orbital datacentres, maybe we should compare notes!
Here's my current draft; I expect perhaps 20-30% of it will need rewriting, not counting anything I've yet to add: https://github.com/BenWheatley/blog/blob/main/2026/06/22-23....
Dotnetrocks did an excellent breakdown https://www.dotnetrocks.com/details/2007
He3 is needed as a cryogenic for quantum systems, not fusion fuel.
Not a coincidence that both technologies aren’t going to live up to the proponents’ claims, and both want materials we don’t really have.
As I understand it, He3 is also used in nuclear materials detectors.
Morgan Stanley list both.
Panhandler

> I sat out the TMT bubble until they quit using the term "information superhighway," and it saved me an 80% drawdown.

> I'm sitting out the AI / chip / SpaceX [AICSX, pronounced like the wrestling shoes?] bubble until they stop using the word "compute" as if it were a noun.

> I'm guessing I'll save myself a drawdown on a similar scale.

A comment under the original article.

> until they stop using the word "compute" as if it were a noun.

That’s a bit silly, since it is a noun at this point, meaning computing resources or computing capacity.

It’s nowhere near being a term similar to “information superhighway”, which was never a technical term used within the industry, it was purely used in communication mostly to the public or in government agency and contractor slide decks.

“Information Superhighway”, ha ha (and replete with billboards as we have come to find out).

As opposed, I suppose, to the library. Which is perhaps more like an information bike lane. (Or a pedestrian walkway?)

That feels…low. They’re the only one with significant proven and operational space lift capacity
Citation:

"Morgan Stanley’s sum-of-the-parts analysis tells a more nuanced story. The “Space” segment, which encompasses Falcon rockets, Dragon capsules, and the Starship program, has been bleeding money. Heavy investment in Starship development drove operating losses in that division, even as SpaceX overall reported a profit of around $8 billion on revenues between $15 and $16 billion in 2025"

In finance and so the world we live in, the value of an equity is less related to the merit of the product than the firm's capacity to generate future free cash flow. Software companies, B2B SaaS in particular, have been basically unbeatable in this regard, hence the state of the market (and our salaries) for the past few decades. Industrial firms have to use metrics like "EBITDA" to show how much cash they'd have to potentially pay to investors if they didn't have to pay so much in interest on their debt, taxes, and fixing up decaying equipment...
But it doesn't make any _money_. Approximately all the ballpoint pens in the world are dependent on a single company, Mikron Group. It has a market capitalisation of about CHF 275.5 million. "We're the only ones who do that" is insufficient; it has to make money. And honestly, generally, "we're the only ones who do that" is a strong sign that a thing doesn't really make much money; if it did, well, other people would do it.
They're treating lift as a cost centre for the $128/share connectivity segment. (X and Grok being worth $12/share is debatable. Enterprise AI being worth $150+ speaks for itself as nonsense.)
There just isn't that much demand for space.

Space is cool to nerds like me, but what do I really need from it? I've got all the navigation satellites I could want (which I don't pay for) and the best satellite imagery I use is still hyperspectral airborne imagery.

Now, of course that's not the full story but the use cases get rather specific beyond that: the launch market just isn't actually very big (afaik $30 billion a year).

Without doing a google search - optical fibers factories and pharma factories can deliver higher quality products when built in space. And I bet there are hundreds of other examples.

Just because launch costs were high and these weren't viable before, doesn't mean they won't be viable now.

Space lift is not very profitable. The money comes from whatever you actually put in orbit with that lift capacity.
SpaceX’s launch services have huge margins.
Starship is going to be huge when it's finally operational. It seems like it should be worth more than $8/shr if the potential is accounted properly.
Why is it going to be huge? Who are the customers and what do they want?

I don't follow this closely, just look at the pretty pictures. If there's demand for lifting much bigger/heavier things to orbit than presently possible, I would probably not know, for lack of pretty pictures. So please tell.

Don’t forget that SpaceX has 3.884 billion outstanding shares!
just like full self driving tm
> Starship is going to be huge when it's finally operational.

Any day now. Yep, real soon, honest!

I'm thinking of flying over on the Spruce Goose to watch the first operational launch.
This feels like the “there’s a global market for maybe 4 computers” of the 21st century.
This feels like the “we should make an internet company that ships pet food to anyone, anywhere!” of the 21st century.
I have a hard time imagining how spx stock will drop to a reasonable price. You have to be a true believer at its current valuation, so who is going to sell? Maybe if enough insiders or stock options start selling it will drop.
Slowly at first, then quickly.
The market overall drops and large investors start shedding their pricier/riskier stocks to offset losses.
I expect via court order or government intervention.

How many of the HODLers will be SpaceX believers, vs. Musk believers? Musk's already only narrowly avoided being banned from running publicly traded companies from the $420 tweet years back.

Well, as I understand it, SpaceX intends to continue to raise money from the market. Eventually the true believers will stop having money to give them.
I can't access the article but it's my understanding the AI bit of SpaceX is a negative valuation (ie losing money like a gutted pig).
It makes zero dollars, but generates a lot of debt. It's worth negative dollars by any sane measure.
I really wish I could buy the SpaceX part of SpaceX decoupled from all the AI hopium.
You might enjoy Rocket Lab $RKLB, pure space tech without the AI distraction.
> In our model, we estimate SpaceX raising an average of $72bn annually between 2027 and 2030 and then an average of $95bn annually between 2031 and 2034

Ah. There it is. Even if that's all done as investment-grade debt (with a 50 bp underwriting spread), that's $3+ billion of banking fees.