Sam Altman spent the weekend telling public market investors that Elon Musk is selling them a short-term story on space data centers, and the engineers who have run the numbers largely agree with him. Responding to Musk calling him a scammer, the OpenAI CEO fired back that Musk is the one pitching orbital compute to retail buyers on a timeline the physics does not support. The exchange matters because SpaceX's roughly $2 trillion valuation leans heavily on the promise that a fleet of orbital data centers will power AI inference at scale.
That thesis is the load-bearing wall under SpaceX's current price. Bullish analysts frame the orbital compute pitch as fuel for SpaceXAI's models or as an in-space neocloud competing with terrestrial hyperscalers. Strip the space-compute narrative out and the multiple compresses fast.
Talk to the people actually building this — founders at rival space data center startups, engineers on Google's orbital compute project, independent analysts running the launch-cost math — and the answer is the same. Scale space compute requires two things that do not yet exist at price: rockets that fly cheaply and repeatedly, and high-powered satellites that can be manufactured en masse.
Key facts
- 01SpaceX's $2 trillion valuation is being driven largely by its orbital AI inference pitch to public market investors.
- 02Starship's 13th test flight is expected as soon as July 16, but operational reusability is still years away.
- 03SpaceX told IPO investors Starship may not be fully reusable near-term and could expend second stages each launch.
- 04Subject-matter experts peg scale manufacturing and launch of orbital data center satellites to the 2030s.
Musk's answer is Starship, the rocket SpaceX is expected to fly for the 13th time as soon as July 16. If the vehicle reaches true operational reusability, the unit economics for orbital data centers could pencil out. But even a clean recovery of both stages on the next test flight does not equal an operational reusable rocket — that is still years of iteration away, and Starship's early cadence will be spoken for by NASA commitments and Starlink buildout well before orbital data centers get their turn on the manifest.
SpaceX itself conceded the issue during its IPO road show, telling investors that Starship may not be fully reusable in the near-term and will need to expend each of its second stages on every flight. Expending the second stage on every launch does not close the business case for economical orbital compute — it eliminates it. That is the unglamorous line buried in the prospectus that Altman's weekend post is effectively pointing at.
“We start flying them next year”— Elon Musk, SpaceX CEO
Musk's rejoinder — that the satellites start flying next year — is technically defensible and strategically thin. SpaceX can absolutely put a satellite equipped for high-speed data processing into orbit in 2026 or 2027 as a demo. The harder question is when the company can build and launch those satellites at the volume required for meaningful inference capacity, and by consensus of the people doing the work, that is a 2030s question.
Physics is the constraint, not ambition. A serious orbital compute business needs multi-megawatt satellites, thermal management in vacuum, ground-station bandwidth to move inference traffic, and a launch cost per kilogram low enough that the whole stack beats a terrestrial data center on total cost of ownership. Each of those problems has a solution path. None of them has a solution on a timeline that supports the near-term revenue narrative baked into today's SpaceX valuation.
The gap between what SpaceX is telling investors and what its engineers are telling each other is the actual story. Space compute is a real long-term business — the roadmap to the 2030s is credible, and SpaceX is the company most likely to execute it. The problem is the timeline mismatch between a public market pricing 2030s outcomes into 2026 revenue and a launch program that still has to demonstrate reusable second stages.
Altman has his own reasons to poke at the pitch. OpenAI's compute demand is enormous, and every dollar the market allocates to orbital compute is a dollar not chasing terrestrial capacity that OpenAI can actually rent this quarter. Framing SpaceX's data center plans as short-term investor theater serves both the truth as most engineers see it and OpenAI's competitive interest — those can coexist.
For the AI market, the read is that the ground game still wins for at least the next five years. Terrestrial hyperscaler buildout — Meta's 5GW Hyperion, the ongoing HBM shortage warnings out of memory suppliers, Sunrun's distributed compute experiments — is where inference capacity gets added between now and the end of the decade. Space compute is real, and eventually it will matter, but investors treating it as a near-term revenue line are pricing a 2030s option as if it ships next quarter.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
