Navy
Modern warships are increasingly poor value because large crewed surface vessels are expensive, highly detectable, and relatively easy to disable, while modern weapons are cheap, precise, and plentiful.
A more robust alternative is a fleet of low cost autonomous hybrid surface–submerged platforms that spend most of their time hidden underwater and surface briefly to launch drones or weapons.
This model only works if autonomy is treated as fundamental rather than auxiliary: these platforms must operate deaf, dumb, and independently for long periods, with surfacing treated as a calculated risk rather than normal operation.
If continuous two way communications are required, the platform should not be underwater, because transmission creates detectability and vulnerability.
One way communications reframes control entirely, with humans dispatching instructions as updates rather than conducting conversations, and with loss of contact treated as normal rather than failure.
Attrition will be tolerated because these AI enabled vessels will be designed to be inexpensive, replaceable, and operationally expendable, shifting naval power from survivable ships to distributed, ambiguous, and software driven disposable launch infrastructure.
Once everyone has these systems, navies stop being about ships and become contests of software, production scale, and loss tolerance, because the ocean fills with cheap, autonomous, deniable launch platforms that no one can reliably defend against.
The advantage goes to whoever can deploy, update, and replace autonomous systems fastest without triggering uncontrolled escalation.
Ultimately war would not even be necessary. The parties could simply model their capability and agree on whatever happens.