With Nvidia on board, SpaceX orbital data centers to hit 'significant scale' in 2028

SpaceX, in partnership with Nvidia, will launch orbital AI data centers using a space-optimized Vera Rubin NVL72 system in Q4 2027, reaching 'significant scale' in 2028, according to Elon Musk. The deployment, which could involve up to 1 million orbital data centers, will use Nvidia Vera CPUs to accelerate agentic AI workloads for the Grok platform and first-generation 'Starmind' AI satellites. Nvidia's Vera CPU is billed as the first built for AI agents, enabling up to 1.8x faster task completion compared to x86 CPUs.

Featured Story ZTE sees network business slump, costs soar and jobs vanish China's ZTE reports a widening delta between revenues and profits as its expensive-to-run enterprise unit powers ahead of networks on sales growth. Using a design powered by Nvidia chips, SpaceX's deployment of orbital AI data centers will start in Q4 2027 and reach 'significant scale' in 2028, predicts Elon Musk. With plans to deploy "up to" 1 million orbital data centers /satellite/musk-s-massive-space-based-data-center-super-scale-or-sheer-folly- optimized to handle AI workloads, Elon Musk has put a general timetable on when he expects deployments to get underway and scale up. In concert with a new agreement https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-vera-cpu-to-accelerate-agentic-ai-at-massive-scale to use Nvidia chips, Musk posted Monday on X https://x.com/elonmusk/status/2091939113008238838 that AI data center launches could get underway as soon as the fourth quarter of next year. "SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028," Musk wrote. "Our design is significantly simpler, lower cost, denser and lighter than a traditional rack," he added https://x.com/elonmusk/status/2091940127866257601 . Musk's timing pronouncement came amid a commitment by SpaceX to deploy Nvidia Vera CPUs to accelerate its next-generation of agentic AI workloads. In addition to expanding AI infrastructure for the Grok platform, SpaceX will also tap Nvidia to support first-generation "Starmind" AI satellites based on an optimized Nvidia Vera Rubin NVL 72 system. Rendering of SpaceX AI1 data center satellite with specs. Source: SpaceX The general aim, Nvidia explained, is for SpaceX to use Vera CPUs to speed up application workloads to enable AI agents to "act faster while keeping GPUs fed and fully utilized." That's all taking shape as expectations grow that agentic AI will alter the shape of network traffic as AI agents talk to other AI agents and require extended sessions. Cisco recently stressed that AI agents generate about 450% more traffic on average /ai-machine-learning/cisco-ai-driving-a-network-supercycle- than a human performing the same task. Nvidia bills Vera as the first CPU built for AI agents, claiming it enables up to 1.8x faster task completion when compared to x86 CPUs. "Agentic AI requires a new kind of computing system – one built not only to generate answers, but to take action," Ian Buck, vice president of hyperscale and high-performance computing at Nvidia, said in a statement. Nvidia, which is scheduled to report Q2 results on Wednesday, said it is optimizing Vera for orbital data centers with respect to power, thermal management and reliability requirements that differ from technology used to power conventional, terrestrial data centers. Of course, SpaceX's orbital data center plan needs a lot more than Nvidia chips. The plan to rapidly deploy AI satellites also hinges on the success of Starship, SpaceX's reusable, heavy load rocket system. In the wake of the most recent test flight of Starship /satellite/starship-completes-first-test-flight-with-starlink-v3-satellites in July, SpaceX plans to execute its first-ever tower catch of the Starship upper stage https://satnews.com/2026/07/25/spacex-targets-first-starship-upper-stage-tower-catch-for-flight-14/ on the next test mission. As for Starmind, the first-gen AI1 model will be 20 meters high with a wingspan of 70 meters wide and support peak power levels of 150 kilowatts and about 120 kilowatts of average compute, and latency in the neighborhood of 3 milliseconds. In June, Musk said the reference design for the AI1 could support both the Nvidia GB300 and the new Vera Rubin platform. SpaceX, he added, will also have a reference design for tensor processing units TPUs , which are chips designed by Google for AI workloads. Others are focused on orbital data center projects, including Google via its "Suncatcher" project , Blue Origin /satellite/bezos-blue-origin-enters-the-data-center-space-race , Jeff Bezos' rocket and lander company, and Nvidia-backed Starcloud. Orbital, a startup that's developing an orbital AI inferencing platform, /satellite/a-startup-called-orbital-joins-the-data-center-space-race recently filed a plan to launch 100,000 satellites /satellite/spacex-and-a-startup-file-separate-plans-to-deploy-100k-satellites .