Jensen Huang used his GTC 2026 keynote to announce NVIDIA’s move into space computing, introducing Vera Rubin Space-1 as the core system in the company’s new platform. The product is designed for satellites and orbital data centers, with the goal of processing data where it is created instead of sending raw information back to Earth first.
Huang framed the idea in a single line: “Intelligence must live wherever data is generated.” In practice, satellites constantly capture Earth imagery and other signals, and sending all of that data to ground stations for processing creates latency and bandwidth constraints. NVIDIA’s answer is to place compute resources directly in orbit.
Dual-chip design targets satellite and station deployments
Vera Rubin Space-1 is built around two chips: IGX Thor for AI inference and Jetson Orin for general computing tasks. NVIDIA said the full system has been redesigned for space requirements, focusing on small size, low weight, and low power consumption. It can be deployed on satellites and could also fit future orbital space stations.
The company also pointed to early application settings through its partners. Axiom Space is building a private space station, while Planet operates a large Earth-observation satellite constellation. Both are direct examples of where space-based edge computing could reduce the burden of transmitting raw data back to Earth.
Heat dissipation, not just compute, is the central challenge
Huang said the main engineering obstacle for a space compute center is thermal management. Data centers on Earth rely on air convection and liquid cooling, but space offers no air, which means heat must be removed through solid-state conduction or thermal radiation.
That constraint shapes the product itself. Huang said Vera Rubin Space-1 will need “quite large heat sinks,” and the cooling architecture has to be redesigned as part of the whole system. For that reason, NVIDIA presented a full space computing platform rather than a standalone chip.