Saab Trains AI Fighter Pilots — By the Billion

Sweden's Saab has built an in-house AI agent suite for the Gripen E, backed by a supercomputer that turns one flight hour into a billion simulated ones.

By The Defense Desk

Saab has unveiled a family of AI agents tested aboard its Gripen E multirole fighter, designed to handle tactical decision-making in combat — with or without a human in the cockpit. [1]

The agents, which use 'various AI techniques to master tactical decision-making in complex operational environments,' can operate alongside a pilot or independently from an uncrewed air vehicle. [1]

Underpinning the program is Saab's AI Fighter Jet Pilot Academy, built to create, train, and evaluate agents. [1] To feed it, Saab has decided to become partial founder in Sferical AI, a Swedish government-backed initiative built on Nvidia's DGX SuperPOD architecture, enabling Large Language Model and Vision-Language Model development for agent-operator communication. [1] The payoff: one hour of simulated flight through Sferical AI produces the equivalent of one billion flight hours for a human pilot. [1]

The infrastructure choice carries a deliberate sovereignty logic. [1] To retain such a capability in Sweden is, he added, 'enormous' for Saab. [1]

The broader ambition is strategic rather than technical: turning a successful flight demonstration into a living capability that adapts as threats do — and doing so without ceding the data that makes the whole system valuable.

Longer term, the AI agents are slated to inform Saab's trio of unmanned aircraft — the A1, A2, and A3 — on the road toward a loyal wingman drone. [1] The academy, in other words, is training pilots for aircraft that don't yet exist.