Broadband telecom is not an easy business to earn a high return in. SpaceX’s revenue is now 55% StarLink—broadband telecom—and 33% “AI”—which means renting out Colossus datacenters to Anthropic—and only 12%, you know, space launches:
Sinead O’Sullivan tells us that “Morningstar, the loneliest bear in the world, puts [the] fair value [of SpaceX] at $780B, of which about… [$60B] launch…[$550B] Starlink”, and $170B xAI. $1.75T of market capitalization on a 2025 revenue of $20B—$$11B StarLink, $6.5B renting Colossus to Anthropic, and $3.5B launches—and a net loss of $5B.
Thus: “The real genius of SpaceX had nothing to do with aerodynamics or rocket telemetry but narrative rotation, as each new iteration of the company’s goals was financed by believers in the last”:
CROSSPOST: SINEAD O’SULLIVAN: SpaceX Pivots Away from Space
<https://www.ft.com/content/0fd15797-ac89-4222-a4bb-03a6a1330f48?segmentID=0bb4f69a-72d2-5619-4945-365b9d6d2888&syn-25a6b1a6=0> <https://www.ft.com/alphaville>
SpaceX pivots away from space
Sinead O’Sullivan
Published YESTERDAY
Sinéad O’Sullivan is a former senior researcher at Harvard Business School and professor in aerospace engineering.
Almost a quarter of a century after SpaceX set out to make humanity multi-planetary, turning the space industry upside down in pursuit of Mars, it appears to have found something better to do.
In most ways, Mars was the entire point of SpaceX — a quasi-religious corporate creed, complete with colonies, timelines and the promise of making humanity a multi-planetary species, carrying our civilisation — and eventually our minds, once digitised — beyond Earth. In February, Musk announced that the company had shifted its focus from Mars to the much closer lunar surface, which also happens to be a more useful place to manufacture orbital data centres. Last week it emerged that SpaceX has stopped bidding for commercial Falcon 9 launches, allowing only US government customers for the rocket. And Gwynne Shotwell signalled last week in Paris that Crew Dragon, the only operational crew vehicle for routine missions for Nasa, will eventually be wound down.
What does it mean for the business of space when a company that is founded on getting humanity off Earth loses interest in the rather more immediate business of keeping humans in orbit around it? Shotwell’s solution? Let Boeing “have some business” — despite Boeing’s failure to fly a single operational crew rotation.
But there’s a bigger question: what happens to the companies that were relying on SpaceX to get to orbit? Musk spent two decades building Frankenstein’s monster; this ecosystem he now seems to regard as too small and boring to bother with. Luckily, there isn’t nearly as much demand for launch capability as Musk had promised. In 2025 Falcon 9 flew 165 times, and just 43 of those flights carried outside customers. Nearly three-quarters of the most celebrated launch cadence in history was SpaceX putting its own Starlink satellites into orbit. Meaning that selling launches to other companies brought in $4.1bn last year, up only 8 per cent in a market that was supposed to have exploded — and most of that came from Nasa and the Pentagon. In the second quarter of this year, space contributed just 12 per cent of SpaceX’s $7.8bn in revenue. Starlink brought in 55 per cent; the AI business, which only arrived in February, already accounted for 33 per cent.
However, Musk founded SpaceX on the bet that there was an enormous market for launch that was shackled by the industry’s high pricing. By making rockets reusable, the cost would collapse and customers would come running for access to low-Earth orbit. In other words, launch demand would be highly price elastic. But when SpaceX removed the price constraint, the expected flood of customers never came. It turns out the old-guard space executives who decided not to spend billions on reusable R&D weren’t all dinosaurs. Some of them may simply have understood the demand curve correctly.
That’s not to say there’s no demand. Yes, there are some customers for launch, and they’re now scrambling for access to orbit. But there just aren’t enough of them to matter to a company valued in the trillions.
Which suggests that the real genius of SpaceX had nothing to do with aerodynamics or rocket telemetry but narrative rotation, as each new iteration of the company’s goals was financed by believers in the last.
Reusability was paid for by Nasa and satellite operators, with federal commercial off-the-shelf and Commercial Crew money providing early proof-of-concept contracts to finance its early rounds.
Starlink was sold to private investors on the back of this reusability.
Mars gave SpaceX much needed urgency; an existential justification of sorts for moving fast and treating rules as impediments to a higher purpose.
And orbital AI compute is now being sold to public markets as the reason to pay almost any price for its stock.
The last rotation into the AI narrative has obviously been the most lucrative of all. In June it listed at $135 a share, a $1.77tn valuation and the largest IPO on record, on 2025 revenue of $18.67bn and a net loss of $4.9bn. That works out at roughly 95 times trailing sales, for a company whose AI segment spent $5.1bn on R&D last year, more than that segment’s entire revenue.
Morningstar, the loneliest bear in the world, puts fair value at $780bn, of which about $611bn is launch and Starlink.
Still, investors are grown-ups, and markets are a dog-eat-dog business.
But the real problem is Europe’s, because for most of the last fifteen years, Brussels has been trying to compete with a company whose launch model has yet to show meaningful viability, at least without putting a Tesla in space, absorbing a social media company and generating cat memes at industrial scale. For the last decade, Europe has attempted to pour money into new launchers designed to match SpaceX’s financing and operational style — without ever asking where the customers are supposed to come from. “Build it and they will come” did not work for SpaceX, whose rockets were eventually paid for by SpaceX’s own satellites. Without a launch company generating its own enormous demand, the numbers are unlikely to work.
Europe has spent years, and enormous amounts of money and institutional energy, trying to build a European SpaceX. But what if it has been competing with a mirage — and in the process, mistaking SpaceX’s dizzying growth for evidence of a dazzlingly large launch market? Which brings us back to the old, boring model. It kills me to say it, but maybe it was the sustainable one all along. Ariane, anyone?
<https://www.ft.com/content/0fd15797-ac89-4222-a4bb-03a6a1330f48?segmentID=0bb4f69a-72d2-5619-4945-365b9d6d2888&syn-25a6b1a6=0> <https://www.ft.com/alphaville>
Brad DeLong here: What do I think of this? Well, is satellite broadband likely to be super-competitive in the long-run vis-à-vis earth-based fiber- and tower-based broadband? Maybe. Perhaps. In the geographically dispersed low population density 20% of the market. But as soon as people cluster in cities the space-basing loses its edge. Are orbital datacenters likely to be super-competitive in the long-run vis-à-vis earth-based ones? I just do not see it. The orbital-datacenter case seems to me considerably weaker than the satellite-broadband one, for which there is genuine physics for serving dispersed customers. Yes, energy is nearly free in space once you have the solar panels, but heat dissipation is nearly free on earth.
Even Morningstar seems to be bending over backwards to burnish the Muskopoly unduly, IMHO.
Only gross innumeracy can make "orbital data centres" seem even remotely viable. We are ruled by an aristocracy of morons.
This raises an interesting question. Is AI going to follow the same trajectory as reusable space launch?
There was a lot of money spent developing reusable rockets, a long term dream technology, but even as capabilities improved and costs dropped, the market did not expand enough to support the industry if the most successful corporation is moving on.
We're seeing a lot of money being spent on AI, yet another long term dream technology. Capabilities have improved and we're seeing costs drop, but it is unclear if the market will expand enough to support the industry.
I think there are some differences, but the big money app for AI is advertising which is why Google, Meta, Microsoft, Oracle and even Apple are as invested as they are. There's only so much money in replacing phone trees and redoing customer Q&A. Applications in manufacturing, operations and research are going to require more highly tuned systems and form an important niche, but won't provide the required returns on a trillion dollar investment.