Palantir and Ondas Launch Regna AI for Sovereign Edge Defense
Palantir and Ondas just spun up Regna AI to give governments full control over defense models that run on edge hardware without internet.
Business · Source: Ondas Investor Relations
What happened
Ondas and Palantir just created a brand new AI company called Regna AI. It is a standalone private company bootstrapped and initially wholly owned by Ondas. The goal is simple. They want to help governments and enterprises deploy fine-tuned, mission-specialized AI models inside their own sovereign environments. Palantir Chief Executive Officer Alex Karp stated that sovereign AI outperforms frontier models while removing dependence on them.
Ondas is stepping up as the first customer. They will use Regna AI models for autonomous mission orchestration at the edge. These models are built to run on hardware like drones, stratospheric balloons, ground vehicles, and surface vessels. They operate even in denied, degraded, intermittent, or limited connectivity zones. The system combines open-weight models with customer-controlled mission data and Palantir operational platforms.
Palantir provides the model-training infrastructure and delivery support. The two companies will collaborate on forward-deployed engineering to tailor capabilities to specific customer workflows. The AI assesses evidence, compares options, and proposes mission plan updates. However, a human operator always makes the final call for authorization. Ondas Chief Executive Officer Eric Brock noted that no government should have to build its AI strategy on models or data infrastructure it does not control.
Key facts
- Regna AI — New standalone private company formed by Ondas and Palantir
- Hundreds of millions to billions of dollars — Estimated cost to build a model from scratch that Regna AI aims to avoid
- October 12, 2026 — Date Ondas and Palantir announced the formation of Regna AI
Why it matters
Sovereign AI is becoming the defining requirement for defense technology. Governments and defense partners refuse to send highly classified mission data across commercial security boundaries. Until now, the choice was brutal. They could either rely on public frontier models and risk data exposure, or spend hundreds of millions to billions of dollars building a model from scratch. Regna AI bridges this gap. Builders targeting defense must now design for local autonomy, strict data residency, and offline edge inference by default.
The AI race is no longer just company versus company. It is country versus country. Sovereign AI means a nation owns the models, the infrastructure, and the final decisions. Companies pursuing these edge AI architectures face a massive trade-off between local autonomy and operational complexity. Fine-tuned models require controlled update paths, hardware-specific optimization, and rigorous audit trails. As this architecture scales from defense into critical infrastructure and industrial markets, expect a massive shift away from centralized cloud APIs toward highly specialized, air-gapped edge deployments.
For builders
Build for constrained edge environments
Defense clients will not rely on constant internet. You must optimize models to run locally on drones and ground vehicles under denied or intermittent connectivity. The government pays for reliability in the field.
Design human-in-the-loop workflows
Autonomous AI is a liability in warfare. Your systems must assess evidence and propose options, but leave the final authorization to a human operator. Products lacking clear audit trails will lose defense contracts.
Leverage open-weight models locally
Customers want to own their entire stack without spending billions to train frontier models from scratch. Fine-tuning open-weight models on proprietary mission data is the new standard for enterprise and government contracts.
My take
No government should build its AI on technology it does not control. Ondas and Palantir are proving that the future of defense tech is highly specialized, fully air-gapped, and completely sovereign. Ondas eating their own dog food as the first customer is exactly how you test a full loop in public. If you are building AI for the government, stop pitching cloud APIs and start optimizing for the edge.
Original reporting: Ondas Investor Relations. This is my rewrite and opinion.