TL;DR
SpaceX has disclosed Gen3 Starlink specs featuring 250 kW of onboard power, 10 Tbps bidirectional capacity, and an Nvidia Vera Rubin NVL72 AI compute system in orbit, turning each satellite into a flying data center more powerful than the ISS.
What happened
- On October 1, 2026, SpaceX released technical specs for its Gen3 Starlink satellite architecture.
- Each spacecraft carries a 250 kW solar array, a fourfold increase over earlier Starlink generations and more than the ISS (120 to 240 kW).
- Bidirectional bandwidth is rated at 10 Tbps per satellite, with an architectural path to 100-plus Tbps in future iterations.
- Onboard compute runs a SpaceX-designed Nvidia Vera Rubin NVL72 system, enabling orbital AI processing, digital beamforming, and dynamic RF allocation without ground-station routing.
- Elon Musk confirmed the specs directly: "Each sat is 250kW and will have a SpaceX-designed Nvidia Vera Rubin NVL72 computer."
Why it matters
- Edge AI moves to orbit: processing beamforming and routing onboard slashes latency and ground-infrastructure dependency, a structural shift in how satellite networks operate.
- Direct-to-device 5G becomes viable at scale: the higher EIRP from 250 kW enables signal penetration into buildings and dense urban areas, targeting unmodified consumer handsets.
- SpaceX's $17 billion acquisition of EchoStar's AWS-4 and H-Block spectrum now has a hardware platform capable of monetizing those licenses as a non-terrestrial 5G carrier.
- Competitors (OneWeb, Amazon Kuiper, AST SpaceMobile) have no publicly disclosed equivalent in power density or onboard compute, widening SpaceX's technical moat in broadband and D2D markets.
- Nvidia gains a new high-margin, high-visibility deployment vector for Vera Rubin silicon, validating space as a credible edge-compute market.
What to watch next
- Starship launch cadence: Gen3 deployment is gated entirely on Starship's ability to lift heavier spacecraft; any Starship schedule slip directly delays commercial rollout.
- FCC and international spectrum approvals: modified orbital authorizations and spectrum coordination are required before service launch; regulatory friction here is the most likely near-term bottleneck.
- Thermal and power validation in orbit: initial test missions will stress-test the Nvidia compute payload's heat management and inter-satellite laser link throughput, the two least-proven subsystems at this power level.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.