SpaceX and Nvidia are taking AI computing beyond traditional data centres. The two companies have announced plans to adapt Nvidia’s Vera Rubin NVL72 architecture for use in SpaceX’s planned Starmind satellite constellation. The project is aimed at putting high-performance AI computing infrastructure directly into orbit.
Read More | NVIDIA, Noetra to Build World’s First National AI Factory for Physical AI in Japan
The first two prototype satellites, known as AI1, are expected to launch in early 2027. If testing goes as planned, wider deployment could begin in 2028. The announcement follows comments from SpaceX CEO Elon Musk earlier this month, when he said the company sees Nvidia’s Vera Rubin architecture as the best option for its future AI infrastructure.
What is Nvidia Vera Rubin?
Nvidia’s Vera Rubin NVL72 is designed as a complete AI computing rack rather than a standalone GPU system. The platform combines 72 Rubin GPUs with 36 Vera CPUs, connected through Nvidia’s sixth-generation NVLink technology. This allows the components to work together as a shared computing system.
Read More | NVIDIA, OpenAI in Talks Over $250 Billion Data Centre Financing Deal: Report
Nvidia says the rack can deliver up to 3.6 exaflops of NVFP4 inference performance and provide 20.7 terabytes of pooled HBM4 memory. The company also claims that Vera Rubin can deliver up to 10 times the performance per watt of its previous-generation Blackwell systems. Space deployment requires a different approach. Hardware designed for a terrestrial data centre cannot simply be placed on a satellite and sent into orbit.
How will AI computing work in space?
SpaceX and Nvidia are developing a modified system called the Space-1 Vera Rubin Module for orbital applications. According to Nvidia, the system can provide up to 25 times more AI compute per GPU than the H100 for space-based inference.
The hardware needs to deal with several challenges that do not exist in conventional data centres. These include limited power, radiation exposure, heat management, communications bandwidth and strict size and weight constraints. Heat is a particularly interesting problem.
Read More | ‘100% Liquid Cooling’: NVIDIA Unveils Technology That Could Nearly Eliminate Data Centre Water Use
Space is extremely cold, but that does not mean a satellite can simply dump heat into its surroundings. In a vacuum, there is no air to carry heat away through conventional cooling. Satellites instead have to radiate heat, which makes thermal design a major part of the system.
Why put AI data centres in orbit?
The idea is driven partly by the growing demands of AI. Large AI data centres require huge amounts of electricity, land and cooling infrastructure. Building new facilities can also involve lengthy permitting processes and negotiations over power and water supplies. Orbit offers a different environment.
Satellites operating in suitable sun-synchronous orbits can receive sunlight for much of their operating cycle, creating an opportunity to use solar power for computing. At the same time, satellite-based AI systems could process data closer to where it is collected.
That could be useful for applications such as Earth observation, autonomous systems, maritime operations and defence. Instead of sending large volumes of raw data from a satellite to Earth for processing, some of the computation could happen in orbit. Only the useful results would then need to be transmitted to the ground.
Could orbital AI reduce latency?
Potentially, but the answer depends heavily on the application. For some workloads, satellite networks could provide alternative routes between locations that currently depend on terrestrial infrastructure. This could become useful where ground networks are congested or where data is generated far from traditional data centres.
Read More | NVIDIA-Backed Startup Offers Homeowners Up to $22,000 a Year to Host AI Data Centres
The bigger opportunity may be edge computing in space. Earth-observation satellites, for example, can collect enormous amounts of imagery. Processing that information on the satellite could allow the system to identify relevant images or events before sending data back to Earth.
That would reduce the amount of information that needs to be transmitted and could speed up decisions. For everyday AI users, however, the change would be far less obvious. A chatbot or AI application would not necessarily reveal whether a particular workload was processed in a terrestrial data centre or by computing infrastructure orbiting hundreds of kilometres above Earth.
Indrani Priyadarshini is a journalist and editorial professional specialising in technology, artificial intelligence, smart cities, green energy, and digital transformation. With over four years of experience in tech journalism and digital media, she is known for turning complex industry developments into clear, engaging, and insightful stories. Her expertise spans reporting, editorial strategy, digital publishing workflows, and in-depth coverage of emerging technologies shaping the future.
