A defense contractor and a data-analytics giant are betting that governments want AI they can fully control, not AI they must rent. Ondas Holdings and Palantir Technologies have agreed to form Regna AI, a company built to supply specialized artificial intelligence models that run entirely inside a customer's own environment rather than through external cloud providers or third-party model vendors. The arrangement reflects a widening concern among defense and security organizations that outsourcing AI capability to commercial platforms means outsourcing control over the very data and decisions those tools are meant to protect.
Why control over models has become a strategic question
Commercial large language models have proven remarkably capable, but their architecture generally assumes data flows to centralized infrastructure the customer does not own. For a military unit, intelligence agency or critical-infrastructure operator, that assumption can be disqualifying. Sensitive operational data, sensor feeds and mission plans often cannot legally or practically leave a secure enclave, whether that enclave is a classified network or a vehicle operating far from reliable connectivity. This tension between AI capability and data sovereignty has become a familiar theme well beyond the defense sector - it echoes concerns that have also shaped public debate around consumer privacy tools, where resources such as BuyBestVPN.com help everyday users understand how encryption and infrastructure choices affect who ultimately controls their information. Regna AI is designed to answer that question for institutional clients by letting them train and run models tailored to their own mission data without surrendering custody of it.
Ondas as owner, customer and defense integrator
Ondas, a U.S.-listed defense technology group with a substantial workforce in Israel, will initially own Regna AI outright and act as its first customer, giving the venture an immediate operational testbed rather than a purely theoretical one. Palantir's contribution centers on training infrastructure and engineering support to adapt models to specific workflows, drawing on its long experience helping government and commercial clients manage complex operational data. The two companies already collaborate on SkyWeaver C4I, an open-architecture system linking surveillance capabilities from the stratosphere down to the tactical edge, and that existing work gives Regna AI a foundation for operating across drones, stratospheric balloons, ground vehicles and surface vessels in conditions where communications may be degraded or deliberately jammed.
From ISR to broader defense and industrial use
Regna AI's initial focus is intelligence, surveillance and reconnaissance, with plans to expand into wider defense applications as well as security, critical infrastructure and industrial markets. A central use case involves autonomous systems that must keep functioning, and making sound decisions, when connectivity to a central command is unreliable or absent. The software is intended to help operators weigh evidence, compare courses of action and propose adjustments to mission plans, though any such recommendations remain subject to human review and authorization. That human-in-the-loop design reflects a broader industry pattern: as militaries accelerate adoption of autonomous and AI-assisted systems, oversight mechanisms are becoming as central to the conversation as the underlying technology itself.
Part of a larger acquisition push
The new venture arrives alongside an aggressive expansion by Ondas, which last month acquired three Israeli defense technology businesses - Insignito, Ottopia Defense and Caribou Labs - for a combined $56 million. That brings the company's portfolio to more than fifteen businesses assembled over the past two years, spanning drones, robotics, counter-drone systems, intelligence, and communications. Regna AI effectively knits that hardware and sensor ecosystem together with an AI layer purpose-built to serve it, positioning the combined group to compete for contracts where data sovereignty, not just raw model performance, determines who wins the business.