Overview
- Palantir and Nvidia formally launched an integrated "intelligent engine" on June 29 that embeds Nvidia’s open-weight Nemotron models inside Palantir’s AIP, Foundry, Ontology and Apollo platforms for air-gapped and classified environments.
- Palantir CEO Alex Karp said multiple U.S. government clients are already running Nemotron through the Palantir stack, including users that support critical infrastructure, but the companies have not named agencies or disclosed contract terms.
- The package lets agencies host, inspect, fine-tune and erase models and data on local hardware so model weights never leave agency control, which avoids token-based pricing and vendor dependence tied to closed cloud APIs.
- Deploying the stack requires agencies to buy Nvidia GPUs, staff engineers for model lifecycle work, and build governance processes to manage telemetry, post-training and ongoing alignment of models in sensitive settings.
- The partnership builds on an October 2025 sovereign-AI reference architecture and positions Palantir and Nvidia as an alternative to hyperscale cloud contracts while raising questions about scale, cost and which agencies will adopt the system next.