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

  "Starting with the reusable rockets, MS expects SpaceX to eat itself. Launch costs will collapse by more than 99 per cent versus their historical average within 10 years, it says. Operating margin on launches will have jumped to about 40 per cent, from around negative 50 per cent currently, but 40 per cent of less than 1 per cent still isn’t very much."

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).

This feels like the “there’s a global market for maybe 4 computers” of the 21st century.

> 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.

I really wish I could buy the SpaceX part of SpaceX decoupled from all the AI hopium.

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.