TL;DR
The FCC has formally accepted SpaceX's "Starmind" filing for 100,000 AI computing satellites, opening the plan to public review and marking the first regulatory milestone for what could become a 1-million-satellite orbital supercomputer.
What happened
- On September 18, 2026, the FCC's Licensing Division accepted SpaceX's Starmind application as an official filing, triggering a public comment period.
- The filing covers 100,000 AI-focused satellites as the first tranche of a broader plan for up to 1 million craft.
- Elon Musk confirmed on September 20th that each satellite carries a SpaceX-designed Nvidia Vera Rubin NVL72 computer, 250 kW of power, and a target of 10 Tb/s connectivity in both directions, with a path to 100+ Tb/s.
- Satellites will orbit at 323-327.5 km and 473-477.5 km altitude, weigh up to 4,000 kg each, and span a 75-meter wingspan.
- Acceptance is not approval: the FCC noted review priority may depend on earlier applications in the queue.
Why it matters
- Orbital AI compute at this scale is unprecedented: 100,000 NVL72-class nodes in low Earth orbit would represent a distributed supercomputing layer with no terrestrial equivalent in footprint.
- The 250 kW per satellite power budget signals these are not communications relays with side-compute; they are purpose-built AI inference and training platforms that happen to transmit data.
- Companion Starlink V3 satellites in the same filing offer up to 1 Tb/s downlink and six 400 Gb/s laser crosslinks, meaning the network backbone can feed the compute nodes at scale.
- Starship's launch economics are central: one Starship manifest delivers roughly 20 times the capacity of a Falcon 9 V2 run, making the mass and volume of 4,000 kg satellites financially plausible.
- Competitors in cloud AI (Microsoft, Google, Amazon) and rival satellite operators face a potential adversary that combines connectivity, compute, and a captive launch vehicle under one roof.
What to watch next
- FCC comment period responses: pushback from incumbent satellite operators, radio-frequency interference objections, or orbital debris concerns could slow or reshape the license.
- September 28 Starship launch: performance data will signal whether the vehicle can reliably handle the mass and cadence required to populate even the first tranche.
- Nvidia Vera Rubin NVL72 supply commitments: confirmation of a dedicated SpaceX production line or priority allocation would validate the hardware timeline and reveal how tightly Nvidia is coupled to this program.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.