NVIDIA announces Space-1 Vera Rubin Module for orbital data centers

What's new

Alongside the GTC 2026 platform reveal, NVIDIA announced a Space Computing line on March 16, headlined by the Space-1 Vera Rubin Module, a flight-rated package of Rubin-class silicon aimed at orbital data centers and on-spacecraft inference. The Space-1 module integrates IGX Thor and Jetson Orin alongside Vera Rubin compute, and NVIDIA claims up to 25x the AI compute of an H100 in a comparable on-orbit envelope. The IGX Thor and Jetson Orin platforms, plus an RTX PRO 6000 Blackwell Server Edition variant rated for spacecraft, are listed as available immediately; the Space-1 Vera Rubin Module itself is positioned for later availability without a public date. NVIDIA named Aetherflux, Axiom Space, Kepler Communications, Planet Labs, Sophia Space, and Starcloud as launch partners. Source: NVIDIA Newsroom, "NVIDIA Launches Space Computing, Rocketing AI Into Orbit."

Why it matters

Orbital data centers have been a research-paper category for years; Space-1 is the first time a major vendor has shipped a productized Rubin-class compute module specifically engineered for the radiation, thermal, and power-budget regime of low Earth orbit. The 2026 deployment volume will be small, but the precedent is meaningful: NVIDIA is committing the same silicon family that runs terrestrial AI factories to a flight-rated SKU, which gives spacecraft operators a plausible path off bespoke FPGAs and rad-hard MCUs for inference workloads. The lineup also reads as a deliberate fit with the geospatial-intelligence and autonomous-rendezvous workloads the named partners are flying, where on-orbit inference shortens the downlink round-trip and reduces the volume of raw imagery that has to be relayed to ground stations.

Caveats

The "up to 25x the AI compute of an H100" figure is NVIDIA first-party and tied to an unspecified workload mix; no independent benchmark or sustained-throughput data has been published. Space-1 has no general availability date, no public pricing, and no disclosed total-ionizing-dose or single-event-upset rating, all of which matter to spacecraft integrators well before peak FLOPS does. Whether the module is rad-hard at the silicon level or rad-tolerant through architectural shielding has not been clarified, and that distinction governs which orbits the part is realistically usable in. Several of the named launch partners have not yet flown commercial payloads at scale, so the early deployment record will be determined by a small number of missions. Source: NVIDIA Newsroom, March 16, 2026.