This time, anyone trekking all the way past Hall 11 found a surprisingly slim spread of dedicated naval exhibitors. Or at least, that’s how it looked at first glance. In reality, the news wasn’t to be found in the Naval Zone, but rather scattered across the show floor. And it wasn’t predominantly about hulls or platforms, as one would expect. It was about the "invisible": AI and autonomy.

For instance, on 10 September, HII and Shield AI announced a partnership to merge the latter’s Hivemind mission autonomy software with the shipbuilder’s Odyssey suite. Hivemind is designed to let uncrewed systems operate and execute complex missions in GPS- and comms-denied environments, while Odyssey provides an open-architecture framework for operations... Continues below

This analysis article originally appeared in September's Decisive Edge Naval Warfare Newsletter.

Newsletter Sponsor:

Both companies aim to extend these capabilities further into the maritime space, starting with USVs — where Odyssey has already been proven with US and allied services — and later UUVs like HII’s REMUS line. The promise is for scalable, modular autonomy across domains, although its naval mission specifics remain to be seen.

Beyond such announcements of AI integration in an ever-increasing variety of systems, what really stood out at DSEI were the conversations happening on stand. What does AI actually deliver to crews (ie will it replace them)? Can it be trusted? And as naval platforms add more sensors, how can navies handle the spiralling energy demands of processing and cooling — especially on small UUVs and USVs?

“AI is not a revolution,” Gabriel Rangoni, VP strategy and marketing at Thales, told me. “It’s a disruptive evolution and its real value lies in helping humans deal with problems that can’t be neatly modelled with deterministic maths - situations where there’s too much complexity, too much data, or too much uncertainty."

Schiebel - Leading the unmanned evolution

Meanwhile, David Young, CEO of UK firm Theyr, noted that the company’s T-VOS dynamic optimisation AI, a multi-objective genetic algorithm, synthesises the natural environment through time and space with a vessel's digital model and mission objectives to deliver a Pareto front of optimal voyages. Ultimately delivering up to 15% savings in vessel and mission performance, T-VOS is now utilised widely in commercial shipping and on the US NOMARS Defiant USV.

But approach this only works if humans trust the AI on which their system relies, which is why Thales uses hybrid AI models. Instead of leaving everything to a “black box” neural net, these models combine data-driven learning with symbolic, deterministic logic. This adds two big benefits: explainability (the system can justify its recommendations without spiralling into nonsense after a few “why?” questions), and efficiency (logic shortcuts mean you don’t need large quantities of training data for every possible scenario). 

This is particularly important in defence, where data is often scarce, classified or fragmented, and where operators will not accept an opaque system making high-stakes recommendations.

Explainability is also the focus of Theyr, which is developing layered techniques to make AI more transparent and interpretable to users. “If you can explain a route in plain language, you go a long way towards influencing a captain’s decision,” Young told me. “That’s the aim of XAIM — explainable AI for maritime — giving crews the ability to interact with, and trust, what the system is recommending.”

Another key concern: where does the energy come from to process all this data? And how do you keep servers cool on small platforms?

Hybrid AI helps by applying simpler logic where possible, saving heavy computing only for genuinely complex tasks. But hardware matters too. Leonardo DRS for one is tackling the problem at the infrastructure level with its IcePiercer ruggedised liquid cooling systems. Instead of relying on traditional, inefficient air-cooled computing racks, its immersion tanks submerge computing components in a completely sealed system of circulating synthetic mineral oil, increasing computing capacity while reducing energy wasted on cooling.

IIn commercial data centre tests, a single liquid immersion tank was able to replace up to eight racks, and cut water use almost to zero. These are big gains when most data centres burn up to 40% of their total power just on cooling. For navies, this matters: the solution is compact, resilient against dust, vibration and saltwater, and scalable down to smaller platforms, making high-performance edge processing viable on ships, submarines or even UAVs.

Taken together, the message from DSEI was clear: naval AI is not about science-fiction visions of machine captains. It’s about systems crews can trust, AI that can explain its choices, and hardware that can handle the power and cooling burden of processing data at the edge. Thales, Theyr, and Leonardo DRS each told that story in its own way — less about hype, more about the practical steps needed to make autonomy usable at sea.

Don't want to miss out on future Decisive Edge content? Make sure you are signed up to our email newsletters.