AitopkitPalantir has a new vehicle for defense AI. On October 12, drone-technology company Ondas and Palantir announced the formation of Regna AI — a standalone private company that will build mission-specialized, fine-tuned AI models designed to run entirely inside customer-controlled, sovereign environments.
The structure is deliberate: governments get models and infrastructure they actually control, not rented intelligence. Ondas will be Regna AI's first customer, deploying the models for autonomous mission orchestration across drones, ground vehicles and command nodes — including under denied or degraded connectivity, where cloud-dependent AI can't operate. Palantir contributes model-training infrastructure and forward-deployed engineering support.
Ondas CEO Eric Brock framed it as a sovereignty argument: "No government should have to build its AI strategy on models or data infrastructure it does not control." It is a pitch aimed squarely at defense buyers who want frontier-grade AI capabilities without sending sensitive data or operational control outside their borders.
The move fits a broader pattern: defense AI is shifting from generic models toward small, mission-specific systems that run on-device or in isolated environments. With Regna AI, Palantir and Ondas are betting that sovereign deployment — not just model capability — will be the differentiator governments pay for.
Brock's sovereignty argument is the sharpest version of a pitch now echoing across the defense world. For years, governments bought AI the way enterprises bought software: as a cloud service, with data flowing to someone else's infrastructure. That model is breaking down under the weight of its own logic — no defense ministry wants its operational planning dependent on a foreign company's servers, terms of service, or political weather. Sovereign AI is the industry's answer: frontier-grade capability that lives inside borders the customer controls.
The technical bet underneath Regna AI is equally notable. Instead of one giant general model, the plan is mission-specialized, fine-tuned models that run where the mission happens — including under denied or degraded connectivity, where cloud AI simply cannot function. This is the edge-AI thesis applied to warfare: smaller, focused models that trade raw generality for reliability in the field. It mirrors a broader shift across the AI industry, where the most interesting engineering is moving from "bigger model" to "right-sized model in the right place."
For Palantir, the structure is strategically clever. A standalone company lets it sell sovereign AI to governments that might hesitate to buy directly from a US-listed firm with deep Pentagon ties, while keeping Palantir's infrastructure and engineering at the center of the stack. For Ondas, it turns a drone company into an AI platform story. Whether Regna AI becomes a template — the joint-venture-as-sovereignty-wrapper — will depend on whether the first deployments actually work when the connectivity drops. In defense tech, demos are cheap and field performance is everything.
The timing is not accidental either. With governments worldwide racing to stand up national AI strategies, the vendors who can offer capability plus control — not one or the other — are positioned to win the largest contracts of the decade. Regna AI is Palantir and Ondas planting their flag early in that race. The question now is execution: sovereign AI is easy to announce and hard to deliver, and defense buyers have long memories for systems that fail in the field.