Own the automation roadmap for compute production as its first product manager: you decide which parts of keeping GPU fleets worth billions healthy become software next, across fleet health, repair and RMA, hardware qualification, on-call, facility maintenance, and the asset model, and you defend the order with numbers: machines per operator, time to return to service, pages per failure mode. Live on the floor and the rotation: embed with the production engineers and facility operators who run the fleet, sit the on-call shift, map where shift hours actually go, and turn that map into the roadmap everyone can cite, including the SOP source of truth and training records a hyperscaler customer asked to audit.