SpaceX is planning to build AI data centers using Nvidia's next-generation Vera Rubin architecture, with a goal of reaching 10 gigawatts of capacity by 2027. The move marks a major expansion beyond the company's core space and satellite businesses into the booming AI infrastructure market.
What is Vera Rubin?
Nvidia's Vera Rubin architecture is the successor to the Blackwell platform, expected to debut in 2026 or 2027. Named after the astronomer who discovered dark matter, the architecture is designed to deliver massive performance gains for AI training and inference. While Nvidia hasn't released full specs, Vera Rubin is anticipated to pack more transistors and memory bandwidth than its predecessor, making it a key component for hyperscale data centers.
SpaceX's plan to use Vera Rubin suggests the company is aiming for cutting-edge hardware to power its AI workloads. The company already operates a large satellite network with Starlink and has significant in-house AI needs for autonomous systems, but this data center project appears to be a separate, large-scale commercial venture.
A 10 GW Ambition
Ten gigawatts is an enormous amount of power — roughly equivalent to the output of 10 large nuclear reactors. For context, the world's largest data centers today operate at a few hundred megawatts. SpaceX's target would dwarf existing facilities, potentially making it one of the biggest AI computing clusters on the planet.
Such a facility would require massive investment in energy infrastructure, cooling, and networking. It's not clear yet where SpaceX plans to build these data centers or how they'll secure the necessary power. The company has not announced specific locations or partnerships with utilities.
SpaceX's AI Pivot
SpaceX has been quietly building its AI capabilities for years, using machine learning for Starlink's satellite constellation management and for autonomous landing systems. But this data center plan signals a shift toward offering AI computing as a service, potentially competing with cloud giants like Amazon Web Services, Microsoft Azure, and Google Cloud.
The company's experience with large-scale infrastructure — launching thousands of satellites, building rocket factories, and managing complex logistics — could give it an edge in constructing and operating hyperscale data centers. Still, the AI hardware market is fiercely competitive, and Nvidia's supply constraints have been a recurring issue.
Timeline and Hurdles
SpaceX aims to hit the 10 GW mark by 2027, a tight timeline given the need to procure Vera Rubin GPUs, build facilities, and secure power. Nvidia's product cycles have faced delays in the past, and the Vera Rubin architecture is still in development. Additionally, the global chip shortage and rising energy costs could pose challenges.
Regulatory approvals for such a large power draw will also be a hurdle. SpaceX will need to work with local governments and grid operators to ensure reliable electricity. The company hasn't disclosed whether it plans to use renewable energy or build its own power plants.
For now, the plan remains a bold ambition. SpaceX has not released a detailed roadmap or budget. The next step will likely be site selection and early procurement agreements with Nvidia. Whether the company can pull off a 10 GW AI data center in under four years is an open question — but SpaceX has a track record of pushing boundaries.
