Palantir Technologies and Nvidia are deepening their alliance with a new blueprint for running AI inside the most sensitive corners of the US government. The two companies announced a Sovereign AI Operating System Reference Architecture, or AIOS-RA, a mouthful of an acronym that essentially means: a full-stack AI data center design built specifically for government and enterprise environments where data can never leave the building.
Instead of relying on commercial cloud providers to run AI models, this architecture lets agencies deploy cutting-edge AI on their own hardware, in their own facilities, under their own security rules.
What’s actually in the box
The reference architecture combines Nvidia’s latest Blackwell Ultra GPU systems, configured with 8 GPUs per node, alongside the company’s networking technologies and CUDA-X software libraries. On the software side, Palantir brings its AIP, Foundry, and Apollo platforms, which handle everything from data integration to model deployment and lifecycle management.
Palantir’s Chief Architect Akshay Krishnaswamy framed the offering in operational terms.
