Elon Musk used SpaceX's first earnings call as a newly public company to make a striking announcement: going forward, SpaceX will build its AI infrastructure entirely on Nvidia hardware, dropping any multi-vendor approach it may have used before. Musk singled out Nvidia's upcoming Vera Rubin architecture as the reason, calling it the strongest rack-scale AI computing design currently available.
The comments effectively ended speculation that SpaceX might diversify its chip suppliers the way many hyperscalers have. Musk said the company values its close partnership with Nvidia "on many levels" and confirmed SpaceX expects to receive a significant share of Nvidia's GPU output next year.
From Data Centers to Orbit
The exclusivity deal isn't limited to ground-based computing. Musk revealed that SpaceX plans to launch its first Starmind AI satellite next year, built around an optimized version of Nvidia's Vera Rubin NVL72 system, packing next-generation Rubin GPUs and Vera CPUs into orbital hardware. According to Musk, the same underlying architecture will power both SpaceX's terrestrial data centers and its space-based computing fleet, with the rack-scale design meant to cut down on the heavy cabling that typically complicates AI server deployments.
The Market Reaction
Nvidia shares climbed more than 4% on the news, extending a multi-day winning streak, while SpaceX's own stock dropped more than 10% before recovering some ground. The announcement landed as a clear blow to AMD, which had been pitching its Instinct MI450 chips as an alternative for large-scale AI buyers and had previously supplied some of SpaceX's compute needs — AMD shares fell more than 6% the same day.
The Bigger Picture
SpaceX used the same call to lay out aggressive growth targets, including a push toward $100 billion in annualized revenue by the end of 2026 and a longer-term goal of $1 trillion a year by 2030, with AI compute increasingly central to that plan. The company said it expects to end this year with more than 2 gigawatts of compute capacity, scaling toward roughly 10 gigawatts by the end of next year — numbers that underline just how much of Nvidia's future order book may now run through a single, deeply committed customer.