Palantir, NVIDIA publish 'Sovereign AI Operating System'

Palantir's AIPCon event functioned less as a product launch than as a customer roll call across government and industry. The through-line was "sovereign AI" — the idea that organizations keep their data, infrastructure and fine-tuned models inside their own environments rather than handing them to an outside provider. Co-founder and CEO Alex Karp framed large language models as needing an application layer and operational context to be useful, and said Palantir is applying that approach in classified settings where organizations can run models at lower cost while retaining control of sensitive information.

The most detailed proof point came from NVIDIA. Alex Neefus, a senior director of solution architecture, described a joint supply-chain application called Command Center intended to help NVIDIA manage component allocation across a global manufacturing network supporting systems such as the Grace Blackwell NVL72. NVIDIA's supply chain involves millions of parts from thousands of suppliers, and the company tracks a metric called "Time of Ownership" to reduce idle time for components awaiting other materials. Palantir's Ontology connects the underlying data, while NVIDIA contributes its cuOpt decision-optimization library and Palantir's AIP tools evaluate scenarios such as reduced memory availability or adding a factory. Notably, Neefus said a fine-tuned Nemotron 3.5 Lightning model outperformed NVIDIA's larger Nemotron 3 Ultra for the specialized task and can be retrained on a single GPU in a few hours.

Beyond NVIDIA, the event surfaced a range of deployments: Cisco offering Palantir's cybersecurity Ontology model through its AI Factory; the FAA integrating siloed safety sources for Flight Plan 2026; L3Harris reporting a glass shortage identified in 10 minutes versus a process that might previously have taken about three months. Commercial speakers included USA Today, Hexion, Acrisure and Elmet, with Hexion citing $300 million of EBITDA improvement from a three-year transformation. The common thread is that in each case the organization appears to be keeping decision authority — and data control — in human hands, which is likely the core of Palantir's pitch as it competes for enterprise and public-sector AI budgets.