SpaceX Eyes Orbital AI Compute as Starship Reuse Pushes Costs Lower

SpaceX CFO Bret Johnsen outlined the company’s strategy across reusable launch systems, satellite connectivity and artificial intelligence infrastructure, emphasizing vertical integration as a central operating principle. Speaking at an investor conference, Johnsen said the company’s approach extends from rocket manufacturing and satellite production to customer relationships, terrestrial computing infrastructure and planned orbital computing systems. “We really believe that we are building the infrastructure of the future,” Johnsen said. He said the company adopted vertical integration early in its rocket and spacecraft operations because it could not find supply-chain providers moving at the pace or meeting the quality standards it required. Starship milestones and reuse plans Johnsen described Starship as central to the company’s businesses, including satellite broadband, direct-to-device mobile service and orbital computing. He said Flight 13 delivered demonstration V3 satellites, included a Raptor engine relight and achieved what he called the second stage’s “softest” and most precise ocean splashdown to date. The company recovered the vehicle after the splashdown, allowing engineers to inspect it and assess how the heat shield performed, he said. Those findings are being incorporated into preparations for Flight 14 and subsequent missions. Flight 14, expected later in the month, is set to be a revenue-generating mission carrying production V3 Starlink satellites, according to Johnsen. He added that the company expects attempts to recover both the first and second stages to be significant milestones later in the year. Johnsen said achieving reuse of both Starship stages could materially reduce deployment costs. He cited SpaceX’s experience reusing Falcon 9 boosters, saying the company has reflown more than 500 rocket boosters to space. Orbital and terrestrial AI compute Johnsen said SpaceX is targeting its first orbital-compute satellite launches next year. The systems would largely use the V3 satellite platform being deployed for Starlink, with modified payloads and larger solar arrays, he said. He argued that orbital computing could address power, cooling and permitting constraints facing terrestrial data-center development. While the company will continue expanding terrestrial computing capacity, he said orbital compute is “definitely the future” and may reach cost parity with terrestrial systems as soon as next year, contingent on progress in Starship reusability. For terrestrial compute, Johnsen reiterated a target to deploy between 5 gigawatts and 10 gigawatts next year, following expected year-end capacity of slightly more than 2 gigawatts. He said the company has visibility into power, facilities and permitting because it controls some of the development itself. Johnsen also said SpaceX had signed another hosting deal earlier in the month that is expected to generate about $1.11 billion per month beginning Dec. 1, or roughly $13 billion in annual recurring revenue. That agreement increased management’s conviction in its target of reaching $100 billion in annual recurring revenue by the end of the year, he said. The company is NVIDIA-exclusive for compute infrastructure, Johnsen said, and expects its relationship with the chipmaker to support GPU allocation. He said SpaceX’s hosting arrangements have generally been structured as 90-day agreements with 90-day termination provisions, reflecting the company’s desire to preserve capacity for its own AI products over time.