AINEWS
In its first earnings report, SpaceX booked $15.83B for AI chips, twice its quarterly revenue, while Elon Musk declared the company 'exclusive to Nvidia' on the call.
SpaceX spent $15.83 billion on AI infrastructure in the second quarter against $7.81 billion in revenue, a roughly 2.0x capex-to-revenue ratio on a single line item, in the rocket company's first report as a public company (Forbes, Jon Markman). Net loss for the quarter narrowed to $541 million as revenue rose 92% year over year, about 13% ahead of analyst modeling. Total capex for the quarter was $18.37 billion, with the $15.83 billion earmarked for AI running roughly $2.6 billion above the consensus estimate closer to $13.22 billion. The only number investors marked was the one Musk had to read aloud.
On the same call, the chief executive named the supplier and locked the door. "We think the Vera Rubin architecture is the best architecture… we're exclusive to Nvidia," Musk said, referring to Nvidia's flagship AI accelerator platform, named after the astronomer Vera Rubin (Forbes, Jon Markman). No second source was named. No volume discount was disclosed. The buyer, on its largest budget line, declared its own single-source dependence on the record. That makes SpaceX unusual among large AI infrastructure buyers, most of whom run multi-vendor procurement to keep supplier leverage in check.
Hours before the call, SpaceX and Nvidia announced the first ship in that supply chain: Starmind AI1, the first satellite in a planned constellation built to run AI inference and training workloads in orbit (Forbes, Jon Markman). Each satellite carries Nvidia Rubin GPUs and Vera CPUs. The license paperwork is already on file: the FCC application SAT-LOA-20260108-00016 and the authorization order DA-26-113A1 are the regulatory anchors for the constellation. A satellite-based AI cluster is a thermal and power engineering exercise: a data center has the gravity, the cooling, and the grid; a satellite has to dissipate the heat itself and harvest its own power from the sun, which is why the choice of accelerator matters more in orbit than on the ground.
The capex figure made the wire rounds as a number ($15.83 billion vs. a consensus closer to $13.22 billion per the same reporting); the exclusivity line made the rounds as a quote. The two together describe a single decision: SpaceX is committing roughly two quarters of revenue to a single vendor's accelerator line, on a hardware roadmap that Nvidia owns end to end. The upside and the risk of that decision run to the same party.
The second-act proof arrived in the S-1. SpaceX is also building Terafab, a planned $20–25 billion chip fabrication complex in Austin, shared across SpaceX, Tesla, and xAI, running on Intel's 14A process, an upcoming chip manufacturing node, with about 80% of output earmarked for space-based data centers (Forbes, Jon Markman, April 2026). The line items in the SpaceX S-1 filing and the subsequent FWP describe the company's stated plan rather than completed capacity. Read against the exclusivity line, Terafab is the bridge from a single-source commitment to vertical integration: SpaceX is buying Nvidia's top silicon at the high end of the cycle and building its own fab to backstop the demand curve on its own balance sheet. A counter-read is that the fab gives SpaceX an exit, so the exclusivity line could be read as opening language rather than a lock-in. The test for that read would be a same-quarter disclosure of a second supplier, or a pullback on the Intel 14A commitment; neither has appeared.
SpaceX did not give full-year capex guidance on a quarter where the AI line alone ran roughly two times quarterly revenue. Asked to repeat the number for the year, management declined (Forbes, Jon Markman).
The tape priced that omission before the next earnings call. SpaceX shares closed the regular session up 9.4% at $125.33, then gave back roughly 7% after hours once the capex figure was read (Forbes, Jon Markman). Nvidia traded above $219, up about 3.5% on the day, with a market value near $5.35 trillion. The day session priced the announcement; the after-hours session priced the spending. The Starmind AI1 satellite is scheduled to reach orbit later this year, and the first full year of public capex disclosure will arrive with the Q3 report.