TL;DR
SpaceX has publicly committed to a constellation of up to 1 million Starmind AI satellites and revealed the cluster architecture that makes it work, while simultaneously pressuring every other orbital operator to share collision-avoidance data or risk catastrophic debris events.
What happened
- SpaceX published a Space Safety page on October 6, 2026, formally stating plans for up to 100,000 Starlink broadband satellites and up to 1 million Starmind AI compute satellites.
- Starmind clusters will run roughly 10 satellites in tight formation, connected by approximately 10 terabits of inter-satellite bandwidth, acting as a single coherent compute unit, per SpaceX VP Michael Nicolls.
- Each AI1 satellite spans 70 meters tip-to-tip, weighs up to 4,000 kg (nearly seven times a Starlink V2 Mini), and targets 150 kW peak and 120 kW average compute power.
- Gigasat factory in Bastrop, Texas is designed for mass production, with deployment of thousands of units targeted as early as late 2027.
- The FCC accepted the 1 million satellite filing in February 2026; SpaceX confirmed the Nvidia compute payload partnership in August 2026.
Why it matters
- Scale is unprecedented: 11,000 Starlink satellites are in orbit today; 1 million Starmind units would represent a roughly 90x expansion of SpaceX's total constellation footprint.
- Orbital compute becomes a product: SpaceX already has disclosed compute-leasing deals with Anthropic and Google on terrestrial infrastructure; Starmind extends that model to space, targeting AI workloads that need near-continuous solar power and low-latency global reach.
- Collision risk is the chokepoint: Starlink logged approximately 164,000 collision risks over six months where a third-party maneuver was not announced within four hours, and encountered risks with roughly 650 unique maneuvering satellites in 2026, with only about half sharing data.
- SpaceX is setting norms by force: Its free ephemeris-sharing platform, backed by a Stargaze network of 30,000 optical sensors, returns risk results within one minute. Framing rivals' data-withholding as "counterproductive" is a regulatory and reputational pressure campaign ahead of its own massive expansion.
- Astronomers and environmental groups are already opposing the constellation at the FCC, citing the 70-meter wingspan as a significant brightness hazard for facilities such as the Rubin Observatory.
What to watch next
- First Starmind launch manifest: Watch for a confirmed Falcon 9 or Starship manifest carrying AI1 test units, and whether the late-2027 deployment target holds as the Gigasat factory ramps.
- FCC and ITU regulatory response: Opposition filings from astronomers and debris-risk advocates could force mitigation conditions; watch whether the FCC imposes mass or brightness limits on the 4,000 kg satellites.
- Ephemeris-sharing adoption rate: If major operators (OneWeb, Amazon Kuiper, Jio's proposed 1,650-satellite LEO network) join SpaceX's platform, it validates the safety framework; refusal hardens the political case for mandatory data-sharing rules.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.