Ondas and Palantir Launch Regna AI, a New Company for Sovereign Defense AI

What happened

Ondas Inc. and Palantir Technologies announced on October 12, 2026 that they have formed Regna AI, a standalone private company focused on sovereign AI for defense and critical operations. Regna AI is described as “powered by Palantir and bootstrapped by Ondas,” and is expected to scale with institutional partners. Ondas will be its first customer, deploying fine-tuned models to run autonomous mission orchestration on the edge.

What Regna AI actually does

The idea is straightforward: instead of governments sending sensitive data to a vendor’s cloud model, Regna AI builds mission-specialized, fine-tuned AI models that run inside the customer’s own sovereign environment — the customer’s hardware, the customer’s data, under the customer’s control.

The models are designed for deployment on edge hardware as well as in data centers. The named targets include intelligence, surveillance, and reconnaissance (ISR) systems — drones, stratospheric balloons, ground vehicles, surface vessels — plus command-and-control nodes. Crucially, they are meant to work in denied, degraded, intermittent, or limited (DDIL) connectivity conditions: places where you cannot count on a cloud connection at all.

What the models do, per the announcement: help operators assess evidence, compare options, and propose mission-plan updates — with final decisions left to human review and authorization. That last clause matters; the company is not pitching fully autonomous strike decisions.

The quote that explains the pitch

“No government should have to build its AI strategy on models or data infrastructure it does not control,” said Ondas chairman and CEO Eric Brock. “Regna AI combines open-weight models, customer-controlled mission data, and Palantir’s operational platform to give defense organizations control over the AI that supports their missions.”

Brock named the target market plainly: the U.S. Department of War, allied governments, and the defense and industrial companies that serve them. Open-weight models are part of the recipe — a customer that can inspect and fine-tune the weights is less dependent on the vendor than one renting API access to a black box.

The division of labor

Palantir provides the model-training infrastructure and supports delivery of the fine-tuned models, according to the announcement. Ondas, which makes autonomous systems and defense/security technology, brings the edge deployment side — it runs drones and autonomous platforms that become the first customer deployment. The two also described a forward-deployed engineering collaboration to tailor the capabilities to specific customer workflows. Per reporting from Seeking Alpha, Regna AI is initially wholly owned by Ondas.

Why this matters

Sovereign AI has been building as a trend for two years, but mostly at the level of national compute clusters and government cloud regions. Regna AI pushes the concept one step further down the stack: sovereign AI at the tactical edge, on hardware that may have no connection back home.

This is a direct response to a real constraint. A military drone that needs cloud AI to interpret what it sees is useless the moment its link is jammed. Models small and specialized enough to run on the vehicle itself, fine-tuned on the operator’s own data, change that equation. The trade-off is the one every defense AI program faces: moving intelligence to the edge reduces dependence, but the human-in-the-loop design is what keeps these systems defensible — politically and legally.

It also shows where Palantir’s ambitions now run. The company built its business turning government data into operational software; Regna AI extends that into training and deploying the models themselves, packaged as a standalone company rather than another Palantir product line. For Ondas, a much smaller company, the partnership is a way to attach its autonomous hardware to a serious AI training pipeline without building one itself.

What to watch

The announcement is a company formation, not a shipped product. The questions that decide whether this matters: who signs on as the second and third customers, what the first deployed models can actually do at the edge, and how the human-review design holds up under real operational pressure. Sovereign AI pitches are easy to announce and hard to deliver — the DDIL requirement is the hardest part of the whole proposition, and the part most worth watching.