SpaceX partnered with Nvidia on the Starmind AI1 compute payload, the orbit-optimized Vera Rubin NVL72 design Musk expects to start flying next year; the company told investors this week it will build its compute exclusively on Nvidia hardware from here. Reports suggest SpaceX is in talks for roughly 130,000 acres of Louisiana marshland to expand Starship production and launch. Musk confirmed a hiring push for SpaceX's on- and off-Earth AI supercomputer clusters, the staffing behind 1.4 GW of nameplate compute today and more than 2 GW guided by year-end.

We are live on X this week at 11 AM EST. Follow us on X here.

This week's analysis, "SpaceX's First Public Quarter: AI Segment EBITDA Turns Positive Two Years Ahead of Our Model," reads the August 4 print against the Mach33 SpaceX Gigamodel: the AI segment turned adjusted EBITDA positive at $1.15 billion, a swing from negative $609 million in Q1, where our model carried it negative through 2028. Contract signings got it there, $14.1 billion of Cloud Services Agreements with Google and Anthropic named and $6.7 billion more signed in the first weeks of Q3. First-half AI revenue of $3.4 billion has already run 1.9x our full-year figure, and first-half AI capex of $23.6 billion has passed our full-year number by a third. The segment still loses money from operations, and one quarter of ramp economics on capacity built earlier is not steady state.