Defense contractor Ondas Holdings and data analytics firm Palantir Technologies are forming a new company, Regna AI, built to give governments and businesses a way to deploy artificial intelligence in sensitive operations without handing control of their models, data or infrastructure to outside providers. The venture addresses a problem that has grown more urgent as militaries and critical-infrastructure operators race to adopt AI: the most capable commercial models are often built and hosted by companies whose terms, servers and jurisdictions sit outside a customer's control.
That mismatch matters more than it might first appear. Just as privacy-conscious individuals weigh whether a commercial VPN provider can be trusted with their traffic, or opt for a VPN with a 30-day refund window to test a service before committing their data to it, governments face an analogous calculation at a much larger scale: can they trust a third party with intelligence feeds, troop movements or classified sensor data simply because that provider's AI model is the most advanced on the market. Regna AI is designed to remove that trade-off by letting organizations run specialized models inside their own environments, including on hardware deployed in the field, rather than routing sensitive information through external systems. a VPN with a 30-day refund window
Built for Environments Where Connectivity Cannot Be Assumed
Ondas, a U.S.-listed defense technology group with a substantial engineering presence in Israel, will initially own Regna AI outright and act as its first customer. Palantir will supply the infrastructure used to train the models and work alongside Ondas to adapt the technology to specific operational workflows. The arrangement reflects a broader shift in how governments weigh the benefits of general-purpose AI against the risks of exposing operational data to systems they do not fully control.
The company's first focus is intelligence, surveillance and reconnaissance work, with plans to expand into other defense applications as well as security, critical infrastructure and industrial sectors. A central use case involves autonomous systems operating where communications are degraded, intermittent or deliberately jammed - conditions common in contested military environments. Regna AI's models are meant to run on drones, stratospheric balloons, ground vehicles, surface vessels and command-and-control systems, helping operators assess information and weigh courses of action even when a connection back to central servers cannot be guaranteed.
Human Oversight Remains a Design Requirement
According to the companies, the software is intended to support decision-making rather than replace it: it can suggest updates to mission plans or flag relevant evidence, but those recommendations stay subject to human review and authorization. That design choice reflects a wider consensus forming around military AI, where automation is increasingly accepted for analysis and pattern recognition but kept apart from the final decision to act.
Regna AI builds on an existing collaboration between the two companies around SkyWeaver C4I, an open-architecture system meant to integrate surveillance across the stratosphere and the tactical edge. That prior work gives the new venture an operational base for training and deploying models across multiple platform types in low-connectivity settings - precisely the conditions many of its intended customers already operate in.
Part of a Broader Consolidation Push
The launch follows a period of rapid expansion for Ondas, which employs roughly 1,000 people in Israel and recently acquired three additional defense technology businesses - Insignito, Ottopia Defense and Caribou Labs - for a combined $56 million. Those deals bring the company's acquisitions over the past two years to more than fifteen, spanning drones, robotics, counter-drone systems, intelligence and communications technologies. Regna AI gives that expanding portfolio a shared AI layer, one built explicitly around the principle that sensitive operations should not depend on infrastructure a customer does not own or fully trust.