SpaceX’s AI1: An Orbital Data Center With Google and Anthropic as First Customers
SpaceX has filed an FCC experimental application for AI1, a flat-panel spacecraft built on the Starlink bus designed to run AI inference workloads in low Earth orbit using roughly 1.5 kW of solar power and radiative cooling, with Google and Anthropic listed as the first experiment collaborators. The mission is explicitly a test of on-orbit processing for AI workloads, with a target launch in the second half of 2026.
What AI1 actually is
AI1 is not a communications satellite. It is a flat-panel experimental spacecraft built on SpaceX’s Starlink bus, carrying a dedicated compute payload designed to run AI inference workloads in low Earth orbit. Instead of drawing grid power and pumping coolant, the satellite harnesses solar energy directly and relies on the vacuum and radiative cooling of space to manage heat. It connects to the existing Starlink constellation through optical inter-satellite laser links, processing data on orbit and beaming results to the ground with minimal delay.
The mission is explicitly a test. SpaceX filed for experimental authority with the Federal Communications Commission to evaluate on-orbit processing for AI workloads under real conditions. In the words of the filing, the purpose is "to test on-orbit processing capabilities for artificial intelligence workloads." This is a feasibility experiment, not a finished product.
What the filing specifies
- Form factor: a flat-panel design of roughly 200 kg, built on the Starlink satellite bus and using optical laser links for communication.
- Power: a solar array generating around 1.5 kW dedicated to the onboard compute payload, a small fraction of what a ground facility draws for comparable work.
- Named collaborators: Google LLC and Anthropic PBC are listed as experiment collaborators in the FCC application (SAT-LOA-20260423-00038), meaning Google’s Gemini models and Anthropic’s Claude are positioned to be among the first AI workloads run in orbit.
- Timeline: a target launch window in the second half of 2026, with on-orbit experiments to begin shortly after deployment.
Google chief executive Sundar Pichai publicly confirmed the partnership on X, describing AI1 as "an orbital data center that will run Google’s AI models in space."
The purpose of the experimental authority is to test on-orbit processing capabilities for artificial intelligence workloads. (FCC Experimental Application SAT-LOA-20260423-00038)
Why move compute off the ground
The appeal comes down to three constraints that bind terrestrial AI infrastructure. Power is the first: data centers are colliding with grid limits, and solar collection in orbit is uninterrupted and free of land cost. Cooling is the second: ground facilities consume enormous volumes of water and energy to manage heat, while space offers a different thermal environment. Reach is the third: an orbital compute layer linked to a satellite constellation could deliver low-latency inference to regions where ground data centers are sparse or overloaded.
If the experiment works, the larger vision is a constellation, not a single satellite. SpaceX has signaled interest in scaling orbital compute into a distributed network, effectively an AI processing layer that follows the existing Starlink footprint and runs inference closer to where requests originate.
What open questions does the experiment have to answer
Feasibility is the immediate hurdle, and several issues sit behind it. Latency from satellite passes has to be characterized against the ground round-trip it aims to replace. Hardware reliability in radiation and thermal cycling is unproven at this scale for compute payloads. And the regulatory questions are genuinely novel: data sovereignty becomes murky when processing happens outside any nation’s jurisdiction, and the environmental trade-offs of launch activity have to be weighed against the savings in ground power and water. AI1 is designed to surface these answers, not to settle them.
What does it signal for the industry
For anyone whose work depends on AI models from Google or Anthropic, the direct effect is years away and contingent on the experiment succeeding. The more useful signal is strategic. The largest AI providers are treating compute supply as a frontier problem worth attacking with aerospace engineering, not just with bigger ground campuses. That tells you the demand curve for AI inference is steep enough to justify launching servers into orbit.
The practical implication is patience plus attention. Nothing about your tooling changes because of an experimental satellite. But the fact that Google and Anthropic are willing to test inference in space underlines how seriously the industry takes the limits of ground infrastructure, and where the next round of capacity might come from.
The bottom line
SpaceX’s AI1 satellite is an experiment to run AI inference in low Earth orbit, solar powered, cooled by space, linked through Starlink’s laser mesh, with Google and Anthropic named in the FCC filing as the first collaborators. A late-2026 launch will test whether moving compute off the ground is viable. If it is, the cloud gains an orbital address, and the bottlenecks of power, water, and land that constrain AI today get a route around them. For now, it is a bet on physics and a clear statement about how hard the industry is willing to work to keep feeding the models.
FAQ
What is SpaceX’s AI1 satellite?
AI1 is an experimental flat-panel spacecraft, roughly 200 kg, built on the Starlink bus. It carries a compute payload designed to run AI inference workloads in low Earth orbit, powered by around 1.5 kW of solar energy and cooled by the radiative environment of space, with optical laser links to the Starlink constellation.
Who are the first customers on AI1?
Google LLC and Anthropic PBC are listed as experiment collaborators in the FCC application (SAT-LOA-20260423-00038), with Google’s Gemini models and Anthropic’s Claude positioned to be among the first AI workloads run in orbit. Google CEO Sundar Pichai publicly described AI1 as "an orbital data center that will run Google’s AI models in space."
When will AI1 launch?
SpaceX’s filing targets a launch window in the second half of 2026, with on-orbit experiments to begin shortly after deployment.