SpaceX’s S-1 filing tells two stories about its AI compute strategy, and Dave Friedman’s analysis on May 26 makes the structural argument that they are best read as parallel rather than sequential. The terrestrial story is the revenue story. The orbital story is the moonshot. Both are presented in the filing with conviction.
The terrestrial story is operational reality. SpaceX is spending billions to build out the Colossus 1 and Colossus 2 data centers (the facilities Elon Musk inherited via the xAI acquisition and subsequently SpaceX’s absorption of those assets). The disclosed external customer is Anthropic, with the $1.25 billion-per-month contract running through May 2029, which we covered in detail on May 22. That is roughly $15 billion in annualized revenue from a single customer, with a 90-day mutual exit clause that complicates the headline number but does not change the order of magnitude.
The orbital story is what makes the SpaceX positioning genuinely distinct from CoreWeave, Crusoe, or any other neocloud. Friedman’s piece notes that the S-1 frames orbital data centers (AI compute satellites that would handle inference workloads in low Earth orbit) as the technical evolution of the Starlink connectivity network. The argument is that SpaceX has already solved the hardest physics problems: mass-producing satellites with sufficient compute, deploying them at scale via Falcon 9 and Starship, maintaining inter-satellite links at low latency, and managing power generation and thermal dissipation in orbit. No other entity is close to solving all four simultaneously.
The structural question Friedman raises is whether orbital compute is actually a credible business or a narrative device that helps justify the terrestrial investment. The honest answer is that it is both. The terrestrial buildout produces revenue today; the orbital roadmap signals to public-market investors that SpaceX’s compute strategy has a unique technical moat that scales beyond the commodity datacenter business. That dual framing is structurally what an S-1 needs.
Friedman’s analysis is fair to both sides. He does not dismiss the orbital thesis. The technical building blocks are genuinely in place at SpaceX in a way they are not anywhere else. He also does not overpromise it. The economics of orbital inference (cost per FLOP delivered, latency competitive with terrestrial datacenters, regulatory environment for orbital data processing) remain unproven, and the S-1 does not disclose the unit economics that would let investors assess them.
The structural skepticism worth keeping in mind: SpaceX is preparing an IPO at a valuation that requires public-market investors to believe the orbital thesis is at least partially credible. The bull case for the IPO needs the orbital story to be more than narrative. If post-IPO investors decide the orbital story is signaling rather than substance, the multiple compresses to terrestrial-neocloud comparables (CoreWeave’s roughly $90 billion enterprise value per gigawatt rate), which would imply a SpaceX compute valuation in the tens of billions rather than the hundreds.
The Anthropic relationship is the structural anchor of either thesis. With $1.25 billion per month flowing through the SpaceX compute infrastructure, the terrestrial business has revenue durability. If Anthropic exercises the 90-day exit clause (which it would do if its compute strategy shifts toward Microsoft Maia chips, the deal we covered earlier in May), the terrestrial story compresses sharply.
For procurement teams negotiating compute contracts now, the SpaceX dual-narrative framing has a specific implication: the company has more degrees of strategic freedom than a single-line-of-business compute provider, which can be an advantage (more aggressive pricing, more willingness to take risk on capacity expansion) or a disadvantage (compute strategy could pivot toward orbital priorities that do not align with your terrestrial workloads). Read the contract terms with an eye on which version of SpaceX you are buying from.
Posted by Dave Friedman on davefriedman.substack.com on 2026-05-26.
