SpaceX announces exclusive adoption of NVIDIA's AI computing architecture and plans to deploy space-based data centers.
During a SpaceX earnings call on August 4, following the company's first quarterly earnings release since its IPO, CEO Elon Musk announced that SpaceX has decided to exclusively adopt NVIDIA technology for its AI computing infrastructure. Musk stated that SpaceX considers NVIDIA's Vera Rubin architecture "the best AI computer available today" and plans to begin launching Starmind satellites equipped with NVIDIA's Vera Rubin NVL72 rack-scale systems next year, extending AI computing capabilities into low Earth orbit.
SpaceX and NVIDIA have established a close collaboration across multiple dimensions. On August 4, the two companies announced they would jointly develop the computing payload for the Starmind AI1 satellite—a spacecraft specifically designed to perform AI computing tasks in low Earth orbit—equipped with NVIDIA's Vera CPU and Rubin GPU. Musk noted that the NVL72 system's design offers advantages over traditional standard rack-based architectures. Consequently, SpaceX plans to deploy the system both on the ground and in orbit simultaneously. During the conference call, Musk clearly stated that all future AI data centers built by SpaceX will exclusively use NVIDIA systems.
SpaceX's vision for orbital computing extends far beyond immediate applications. The company envisions ultimately building an orbital network of up to one million satellites interconnected via laser links to enable high-speed data transfer between spacecraft, forming a distributed space-based AI supercomputer. This partnership marks SpaceX's official entry into NVIDIA's space computing ecosystem; when NVIDIA launched its space computing platform in March, SpaceX was not among the six initial launch partners announced.
On computing capacity expansion, Musk has set aggressive targets. SpaceX expects its online computing capacity to exceed 2 gigawatts by year-end and approach 10 gigawatts—rather than the previously anticipated 5 gigawatts—by the end of next year. The company aims to cumulatively reach 20 gigawatts through a series of projects by year-end next year. Musk also noted that current resource allocation is prioritizing power and cooling infrastructure over GPUs themselves. He acknowledged that some projects may not be completed entirely on schedule but expects approximately 15 gigawatts of online computing capacity to be operational by the end of 2027.
Orbital data centers are considered a potential alternative to large terrestrial data centers. Compared with ground-based facilities, space-based deployments require no land use and face relatively less thermal dissipation pressure; however, launching servers into orbit and establishing network connectivity present significant engineering challenges.