Research Feed

How much of the estimated $25.1B per-gigawatt end-user dollar does SpaceX retain as EBITDA? Part 2 of Mach33's pre-IPO AI compute series walks the cost cascade across three business-model configurations and three macro scenarios. Two findings: chips carry the highest intrinsic margin in the stack (67% vs 15% at the model layer), and under bear-case foundation-model conditions, the Anthropic Colossus 1 wholesale lease becomes the most resilient piece of the portfolio.

In this episode of the Mach 33 Private Podcast, Aaron Burnett and Vlad Saigau break down SpaceX’s newly filed S-1 and compare it against their pre-filing model, with most assumptions landing within 5%. They unpack the biggest surprises, including Anthropic’s massive compute deal and the IPO lockup structure, while explaining why declining Starlink ARPU reflects global expansion, why EBITDA matters more than net income for SpaceX today, and how Starship could drive a major reduction in launch costs. They also cover the $20B bridge loan, EchoStar spectrum, NTN chips, Flight 12, and what to watch next.

Episode 4 of the Mach33 Private Podcast breaks down the AI revenue stack and how SpaceX is pursuing three AI business models across infrastructure, models, and orchestration. The episode also covers Grok vs. Claude, orbital compute, Cowboy Space, the EchoStar deal, and why Starship V3 may be the biggest near-term driver of SpaceX’s financial trajectory.

Mach33's bottom-up framework prices one gigawatt of AI compute at $41.6 billion in annual revenue across a five-layer value chain. Applied to SpaceX's three business models: Anthropic wholesale, Cursor hybrid, Grok full-stack, the framework reads the company's progressive internalization of value capture through the 2026–28 buildout.