Centro de Análises de Sistemas Navais (CASNAV), the Brazilian Navy's naval systems analysis centre, began the first installation phase of a high-performance computing system at the Navy Data Centre on 20 August, funded through its Drones Marítimos de Superfície uncrewed surface vessel project1. The cluster holds four NVIDIA H200 graphics processors and storage, for automatic maritime target detection and classification, image processing, and autonomous-navigation support.
Perception models have to be trained somewhere, on imagery of the water a navy actually patrols. Brazil has put that somewhere inside its own facility, which means the trained models and the data behind them survive a change of hull supplier, and a vendor that loses the boat contract does not walk away with the recognition system.
CASNAV's release names no vessel, no trial and no deployment, and describes the spending as research infrastructure2. Other navies bought development capacity first as well: the Royal Navy tendered a £10 million multi-role uncrewed vessel testbed on 5 August. One bought a hull to try things on, the other bought the compute to try things in.
