Palantir and NVIDIA name new customers for their sovereign AI stack
On 21 October 2026, Palantir announced that a new group of customers has started using the sovereign AI offering it builds with NVIDIA. The customers named were L3Harris, McCarthy Building Companies, NVIDIA itself, Ondas and Regna AI, and Tampa General Hospital.
This follows a joint announcement in September, which described how Palantir's Foundry, Ontology and AIP were being used on NVIDIA's own supply chain. According to the October release, the two companies have built what they call an operating system to shorten the path from wafer to deployment. It combines Palantir's AIP platform, NVIDIA's Agent Knowledge Foundation and post-trained NVIDIA Nemotron models. The release also said Ondas and Palantir are working together on manufacturing, mission planning and edge intelligence across Ondas' businesses.
The release didn't include much technical detail, contract values or deployment numbers, so treat it as a direction of travel and not a finished product specification.
Why it matters
The customer list covers defence, construction, healthcare and drones, so this isn't only a government offering. In general, "sovereign AI" means running models and data on infrastructure an organisation or country controls, instead of sending everything to a shared public model. Pairing Palantir's data and workflow layer with open NVIDIA models makes that approach easier to build on.
That's relevant in Australia, where many organisations in health, defence supply chains, mining and critical infrastructure have data-residency requirements, privacy obligations or contract terms that rule out some hosted AI services.
What to do with this
If you already use Foundry, the Ontology is the main asset. Agents and models built on top of it are only as useful as the data model underneath, so tidying up object types, permissions and data lineage now will pay off whichever model you run later.
If you're evaluating Palantir, ask three things in early conversations:
- where the models and data will physically run
- which NVIDIA components are needed and what they cost
- whether a production reference customer exists in your industry
Palantir was scheduled to report its third-quarter results on 2 November (US time). It's worth checking that update for any detail on how widely this work is being adopted.
Sources