SpaceX is sharpening the public profile of its orbital computing ambitions, launching a dedicated online presence for a project that seeks to relocate substantial AI processing capacity off the planet. The initiative, previously identified in regulatory documents as the Orbital Data Center constellation, is now advancing under the Starmind name and is centered on a new satellite platform designated AI1. The updated roadmap envisions a constellation that could eventually scale to as many as one million units, with compatibility for AI silicon from any manufacturer, including chips produced by Nvidia or by Tesla’s Terafab 2 nm foundry.
Inside the AI1 satellite design
Each AI1 satellite is engineered to deliver an average of 120 kW of AI computing power, with peak capability reaching up to 150 kW, placing a single orbital unit in a performance class comparable to a contemporary terrestrial server rack. The physical footprint is considerably larger: the design rises 20 meters tall and, with solar arrays fully deployed, spans 70 meters across—wider than a Boeing 747 fuselage. That scale dictates a reliance on SpaceX’s in-development Starship rocket for launch operations.
Much of that volume is driven by the solar array required to sustain the compute payload. The satellites are planned to operate at an approximate altitude of 600 km, a shell deliberately chosen to avoid interference with the up to 100,000 Starlink Gen3 units—themselves roughly the size of a Boeing 737—that will populate lower orbits.
Orbital inference and thermal management
Unlike second- and third-generation Starlink satellites that relay internet traffic to compact ground terminals, Starmind units are designed to perform AI inference workloads directly on orbit. Results would shuttle between satellites over laser links before being handed off to the Starlink network for the final downlink to ground stations.
The project underscores several structural advantages of a space-based computing platform. Starmind’s public materials point to continuous solar energy availability, natural vacuum conditions, and the elimination of zoning disputes and water-supply constraints that increasingly burden terrestrial AI data centers. Thermal control falls to liquid radiator panels spanning 110 m² that, as described by SpaceX, "reject heat into the vacuum of space." The system incorporates redundant pumped loops and protective shielding designed to guard against micrometeoroids and orbital debris.
Scaling challenges and production timeline
Expert scrutiny has focused on the practicalities of radiative cooling in vacuum and on the broader implications of adding a vast number of AI satellites to an already congested orbital environment. SpaceX has indicated it intends to address these challenges iteratively during deployment and early operational phases.
The current schedule targets the first AI1 prototype flights in early 2027, with volume manufacturing expected to follow later that year from a new production facility called Gigasat. Whether the company can expand the constellation to the million-satellite architecture described in its detailed FCC submission remains an open question, but the brand launch and ample post-IPO capital signal that Starmind has moved decisively beyond initial concept work.
Source: www.spacex.com