Fusion energy design is limited less by concept generation than by the difficulty of integrating many tightly coupled subsystems into a validated plant design. Magnetic confinement fusion requires simultaneous consideration of plasma physics, magnets, structures, heating systems, tritium breeding blankets, and fuel-cycle performance, together with their multiscale and multiphysics interactions. This talk shows how AI can accelerate fusion design across surrogate modeling, workflow automation, and digital twins.

First, AI surrogates for expensive simulations can enable rapid exploration of design trade spaces and support gradient-based optimization through fast, differentiable evaluations. To be reliable, however, such surrogates must be monitored for out-of-domain use and updated as exploration moves beyond their training regime; methods that capture underlying system dynamics may generalize better than purely black-box models. Second, AI agents are becoming capable of orchestrating complex simulation workflows, including meshing, job launch, analysis, and recovery from failed or incomplete runs, thereby reducing the effort required to execute large integrated studies. Finally, the same AI-enabled modeling stack can support digital twins for device operation, where mechanistic models are continuously updated using sensor data to better represent evolving system behavior. More flexible world-model approaches may further enable realistic scenario generation, operational planning, and control. Together, these capabilities suggest a path for AI to accelerate both the design and operation of fusion power plants.