Inference infrastructure expands from centralized clouds to edge and local clusters

ASUS is broadening cloud-to-edge AI hardware, NVIDIA PAIR targets heterogeneous local GPU pooling, AMD is positioning ROCm 10.0 for agent-operated accelerator workflows, and Equinix plans an inference exchange with NVIDIA and Together AI for Q1 2027. Pricing, benchmarked performance, scheduling overhead, capacity, power, and data-sovereignty details remain unresolved.