Navantia and Multiverse Computing pursue embedded naval AI

Spanish shipbuilder Navantia has signed a memorandum of understanding with Multiverse Computing to investigate artificial intelligence and quantum-inspired optimisation for naval defence systems. The partners will assess compact AI models, signal processing, mission planning and command applications, with development and validation needed before any operational installation.

The agreement, announced on 21 October, was signed by Navantia chief operating officer Gonzalo Mateo-Guerrero and Multiverse Computing co-founder and chief executive Enrique Lizaso. It combines Navantia’s naval design, platform integration and in-service support experience with software developed to reduce AI models’ computing and energy requirements.

The parties will examine efficient AI for embedded equipment aboard surface vessels and submarines, together with situational awareness, logistics optimisation and command, control, communications and intelligence systems. They also intend to consider joint proposals for European defence and innovation programmes. No funded trial, named vessel or new customer contract was disclosed.

Multiverse Computing will make its CompactifAI model compression technology available for assessment. Its purpose is to reduce the size and computational cost of AI models while retaining useful performance for selected tasks. Navantia will contribute operational requirements and the engineering constraints arising from the hardware and software installed on naval platforms.

Running AI software aboard a ship can present different requirements from operating a model in a remote data centre. Onboard processors have finite memory, electrical power and cooling capacity, while external connectivity may be limited or unavailable. A smaller model may make local processing feasible in equipment that cannot accommodate larger computing systems.

Model compression changes the computational representation of a system, with the exact techniques and resulting accuracy dependent on the software and task. A reduction in stored parameters or required calculations can lessen demand on processors, but engineering teams must measure whether the modified model still produces satisfactory results under the conditions in which it will be used.

The term quantum-inspired refers to algorithms or mathematical approaches informed by methods associated with quantum computing. It does not mean that a quantum computer is to be installed on a vessel. Neither company has identified quantum hardware as a requirement of the proposed naval collaboration.

Sensor processing is one intended application. Radar, acoustic and other naval sensing equipment can generate substantial streams of information that must be interpreted before they contribute to a recognised operational picture. An AI model might assist with classification or prioritisation, but performance depends on the characteristics of the sensor data and the conditions under which observations are collected.

Noise, incomplete measurements and unfamiliar operating conditions can affect automated outputs. Model evaluation therefore needs to include representative test data, detection of limitations and procedures for reviewing uncertain results. A laboratory demonstration using selected examples does not establish the reliability of a deployed naval capability.

The partners also identify situational awareness and mission planning. These functions involve bringing together information from several sources and evaluating available options within an established command process. The memorandum does not describe autonomous command decisions, changes to rules of engagement or the replacement of personnel responsible for approving operations.

Logistics optimisation provides another possible application because vessel availability depends on maintenance schedules, spare parts, personnel and port or support arrangements. Computational tools can help evaluate plans against known constraints, although the usefulness of an output depends on the accuracy of the information entered and the assumptions used by the model.

Integration with command and intelligence equipment requires attention to interfaces and security. Software must exchange information through approved systems, maintain the required configuration and operate within the available processing capacity. Where equipment is used for defence, technical changes may require additional testing and authorisation before installation.

Naval platforms also have long service lives. A computing system introduced during construction may need to be maintained and updated for many years, while the surrounding sensors and software continue to change. The partners intend to examine how AI models can be supported throughout that life cycle, rather than treating initial model performance as sufficient for deployment.

CompactifAI could reduce the resources needed to run selected applications, but Navantia and Multiverse Computing have not released naval benchmarks for memory savings, inference speed, accuracy or electrical consumption. Those figures cannot be calculated from general marketing claims about model compression, and no shipboard demonstration has been identified.

The cooperation also includes opportunities to pursue European development funding and advance domestic technological capabilities. Such ambitions do not themselves establish a funded European programme or guarantee that every component of a future naval system will be manufactured in Spain.

Further work will centre on selecting candidate applications and testing them against operational, hardware and support requirements. Any resulting installation aboard a surface vessel or submarine would require separate engineering integration and procurement decisions. The October memorandum establishes a development framework, not an operational naval AI product.