SIGNAL
A moonshot operator built infrastructure for its own flagship, watched the flagship fall behind, and rented the spare capacity to its rivals. That is now a repeatable pattern: when a flagship product cannot fill the plant it was designed to serve, the cheapest path to revenue is to lease the underused racks, pads, or turbines to whoever is winning the next cycle. SpaceX is the cleanest live case: a rocket company that built data centers to train Grok, watched Grok lose ground, and sold spare compute to Anthropic in May and Google in June. Whoever controls power, cooling, and rack space now sits one step upstream of the model labs themselves.
The AI segment lost $1.5 billion on $18.37 billion in quarterly capex, and the company still posted a $143 million net loss, because building the factory before signing the tenant is the whole point. Starlink, the only profitable segment, subsidizes the rest at $4.2 billion a quarter. The Verge's read of SpaceX's first post-IPO quarter names the segment map: $2.6 billion in AI revenue against $962 million from launches, with connectivity doing the quiet funding. Musk's investor-call boast about building AI compute capacity at scale faster than anyone else lands beside the $1.5 billion loss, not as a thesis. SpaceX is now a three-segment hybrid: connectivity that prints cash, launches turning into a cost center as Starship absorbs the spending, and a neocloud (a company renting AI compute to other AI companies) paying to learn whether renting to frontier labs is durable or a side bet tied to Grok's recovery.
The pattern repeats wherever ambitious infrastructure lands before the demand meant to fill it. The factory is no longer the product. The lease is.
Reported by Sky for Type0, from SpaceX made more revenue as an AI company than a space company. Read the original: theverge.com