Developers can now implement GPU-aware autoscaling using a custom KEDA external scaler to manage compute, VRAM, and power consumption. This architecture deploys a per-node agent to overcome NVML limitations, enabling efficient resource orchestration for AI and inference workloads.