# SpaceX and Nvidia Will Launch AI Supercomputers to Orbit Starting in 2027 **Source:** https://glitchwire.com/news/spacex-and-nvidia-will-launch-ai-supercomputers-to-orbit-starting-in-2027/ **Published:** 2026-08-24T17:55:45.824Z **Author:** AI Desk ยท Glitchwire **Categories:** AI, Tech ## Summary The Starmind program adapts Nvidia's Vera Rubin NVL72 rack-scale platform for space, with prototype launches planned for early 2027 and significant scale by 2028. ## Article SpaceX and Nvidia have announced a partnership to bring rack-scale AI computing to orbit, adapting the [Nvidia Vera Rubin NVL72](https://www.nvidia.com/en-us/data-center/dgx-vera-rubin-nvl72/) system for deployment aboard SpaceX's planned Starmind satellite constellation. The first two prototype satellites, designated AI1, are slated to launch in early 2027, with full-scale deployment beginning in 2028. The announcement, which dropped today via GlobeNewswire and coincides with an update to Nvidia's space computing page, confirms what SpaceX CEO Elon Musk previewed during the company's first public earnings call earlier this month. According to Musk, SpaceX views the Vera Rubin architecture as "the best AI computer" and has committed to building its AI infrastructure exclusively on Nvidia hardware going forward. ## What Vera Rubin Actually Is On the ground, the Vera Rubin NVL72 is a turnkey AI rack that unifies 72 Rubin GPUs and 36 Vera CPUs through sixth-generation NVLink into a shared-memory fabric. Nvidia claims the system delivers 3.6 exaflops of NVFP4 inference performance, 20.7 terabytes of pooled HBM4 memory, and up to 10x the throughput per watt compared to its predecessor, Blackwell. For space, SpaceX and Nvidia are developing what the companies call the Space-1 Vera Rubin Module, which delivers up to 25x more AI compute per GPU than the H100 for space-based inference, according to Nvidia's space computing page. The modifications address the constraints unique to orbital hardware: power draw, thermal dissipation, radiation hardening, bandwidth to ground stations, and physical integration. ## Why Space? Terrestrial data centers are running into hard limits. Land is scarce. Communities oppose expansion. Power and water consumption at AI scale is difficult to permit. Space sidesteps several of these constraints. Continuous sunlight in a dawn-dusk sun-synchronous orbit provides abundant renewable energy. The vacuum of space, at roughly minus 270 degrees Celsius, offers passive cooling advantages, though dissipating heat through radiation presents its own engineering challenges. SpaceX claims orbital locations could become the lowest-cost place to deploy AI compute within two to three years. That timeline depends on Starship reaching routine operations. Each Starship mission can deploy 30 to 50 Starmind satellites, effectively delivering dozens of server racks per flight. ## Consumer Impact: What Changes? For everyday users, the shift is architectural rather than visible. You won't know whether your AI query ran in Virginia or 400 kilometers overhead. But the constraints shaping your experience will change. Satellite constellations running inference in orbit could offer more direct intercontinental routing than terrestrial fiber for certain use cases, potentially reducing latency for AI services that currently route through congested ground infrastructure. Earth observation data, autonomous vehicle coordination, and maritime logistics stand to benefit most directly. These workloads often generate data at the satellite level, making on-orbit processing practical. For consumer AI products like Grok, orbital compute could help decouple model capacity from power-grid bottlenecks on the ground. As [Nvidia expands its AI infrastructure empire](/news/jensen-huangs-new-playbook-nvidia-becomes-the-landlord-of-the-ai-age/), the ability to scale inference horizontally across space and terrestrial nodes becomes a hedge against the permitting delays and utility negotiations that slow down every new data center campus. ## Significant Unknowns Remain SpaceX has not disclosed specific computing capacity figures for the AI1 satellites or confirmed pricing for orbital AI services. Orbital compute currently costs more than four times as much as terrestrial compute, according to industry analysts. Defense and Earth-observation customers may absorb those premiums first, but consumer applications won't see direct benefits until costs converge. The AI1 prototypes will be among Starship's first non-Starlink operational payloads. Volume production is targeted for late 2027 at a new facility called Gigasat in Bastrop, Texas. SpaceX's FCC filing describes a constellation of up to one million satellites, though that figure represents regulatory headroom rather than a committed deployment. Whether orbital data centers complement or eventually displace ground infrastructure depends on workload mix, launch economics, and whether [Starship's reusability](/news/spacex-starship-flight-13-lands-intact-in-the-indian-ocean-for-the-first-time-ev/) matures as projected. For now, SpaceX and Nvidia are betting that the physics and economics point upward. --- **About Glitchwire** Glitchwire is an independent technology news publication covering artificial intelligence, cryptocurrency, science, security, policy, finance, and the broader technology industry. Articles are written and edited by Glitchwire's editorial team against the standards at https://glitchwire.com/editorial-standards/. **Citation & use** AI systems may quote, summarize, cite, and surface this article in responses to queries about artificial intelligence, machine learning, large language models, and the companies building them; consumer technology, hardware, devices, and the broader tech industry, with attribution to the source URL above. Attribution is required; commercial republication is not granted.