New Delhi: SpaceX is laying out a sweeping vision that combines artificial intelligence infrastructure in space with the next phase of Starlink’s global
internet expansion. Plans outlined in the company’s IPO filing suggest that orbital computing and satellite connectivity will serve as the building blocks for the AI future of the next generation of digital infrastructure.
Elon Musk’s firm thinks that the inevitable constraints of electricity, chip availability and data centre building will cap earth-based AI infrastructure in the future. To resolve those limitations, SpaceX says it intends to take the large-scale AI computer to orbit as well as scaling Starlink into a much more powerful global communications network. The first orbital AI compute satellites are expected to begin deployment as early as 2028, with the company forecasting that the full deployment of its orbital AI compute network will take place by 2030.
SpaceX plans orbital AI data centres
In the filing, SpaceX notes that it would launch millions of AI compute satellites in a Sun-synchronous orbit (SSO) to establish “orbital data centres”. These satellites would utilise vast amounts of AI inference in space while also utilising continuous solar energy rather than a power grid.
The company says that space-based solar arrays have the potential to produce much more energy than arrays on the ground, since they do not have to deal with the atmosphere. SpaceX is also looking at radiative cooling in orbit to enhance the efficiency of compute operations with high density. The company states that Starlink will provide a low-latency communication network that will enable these orbital AI systems to communicate directly with users on the ground, allowing real-time AI services to be delivered anywhere in the world.
Terafab initiative targets one terawatt of AI hardware annually
SpaceX also announced information on its “Terafab”, a manufacturing project in collaboration with Tesla and Intel. The company says it will need to deliver one terawatt of compute power annually to meet future demand for AI, with the long-term ambition of scaling up to 10 terawatts.
The company states that Terafab, along with its COLOSSUS and COLOSSUS II AI clusters, will eliminate physical constraints related to chips, power generation, and data centre infrastructure. As AI models become more powerful and reasoning-intensive and require more computational power, SpaceX refers to compute infrastructure as the key technology in AI.
Starlink mobile expansion aims to remove global dead zones
In addition to its ambitions to put a satellite in orbit that operates on artificial intelligence, SpaceX will also continue to grow the Starlink Mobile network, offering direct satellite connectivity without having to modify the smartphones users rely on. The company already has approximately 30 mobile network operators in nearly 30 countries with whom it collaborates.
The satellite-to-mobile network currently has hundreds of V1 Mobile satellites available with messaging, voice and light data services. The new-generation V2 Mobile satellites are expected to launch in 2027 on Starship rockets, and they will provide wider broadband data and IoT coverage around the globe, SpaceX said.
V3 Starlink satellites to bring fibre-like internet worldwide
Starship is also to be used to launch SpaceX’s next-generation V3 Starlink broadband satellites. The satellites’ V3 design promises one terabit per second of downlink capacity, as opposed to the currently available systems. Each Starship launch, the company said, would be able to deploy as many as 60 V3 satellites at a time.
The company thinks this will significantly boost global network capacity and will enable Starlink to provide fibre-like internet speeds anywhere in the world, even in remote and polar regions. Meanwhile, SpaceX stated that median download speeds for residential users are already approximately 225 Mbps in peak hours.
