Scaling ML Processing with External AI Accelerators