Palantir launches engine to deploy NVIDIA Nemotron open models in sovereign environments

Palantir Technologies announced a strategic initiative with NVIDIA to deliver an intelligent engine for running NVIDIA AI and Nemotron open models in sovereign environments, focused on US government agencies and critical infrastructure. The offering combines NVIDIA's compute, ecosystem, and open models with Palantir's AIP, Ontology, Foundry, and Apollo products, letting organizations train and deploy open models alongside proprietary technology while retaining control of data, IP, and AI systems.

Key capabilities include explicit data authorization, secure perimeter enforcement, architecturally-enforced customer-specific isolation, data portability, right to erasure, and full auditability. Critically, customers own self-improving models specific to their mission: the engine collects user telemetry and trace data, then uses it to post-train and align the model to high-value tasks — a continuous feedback loop that improves operational performance while keeping everything inside the customer's perimeter.

The move captures why open-weight models have become the substrate of choice for national-security and regulated deployments: they can be inspected, fine-tuned, and run entirely air-gapped, unlike closed API models. It complements AWS's parallel GovCloud release of Nemotron and GPT OSS, showing multiple vendors racing to serve sovereign AI demand.

The announcement coincided with a colorful side note: in a CNBC interview about Palantir's NVIDIA deal, CEO Alex Karp reportedly went off-script for roughly 20 minutes to lambaste the AI industry and the US government's relationship with it. Separately, NVIDIA chip rival Etched hit a $5 billion valuation with $1 billion in booked orders for its inference-focused 'frontier inference clusters.' Watch which agencies adopt the Palantir-NVIDIA engine and whether the self-improving feedback loop delivers measurable mission gains.