Reinforcement learning, motion diffusion models, onboard perception, all running across different compute environments. Training these policies runs on GPU clusters. But the character itself runs inference on embedded hardware in real time, navigating guests and generating motion autonomously That edge-to-cloud coordination, simulation and training in the data center, real-time inference on the robot, data flowing between both, is the compute orchestration challenge that scales with every character you deploy. It's the problem #NuNet is solving: a protocol that coordinates workloads across heterogeneous infrastructure, from cloud GPUs to edge devices. The next wave of physical AI isn't bottlenecked by GPU power. It's bottlenecked by the infrastructure layer that ties it all together.