With increasing compute power on the edge, the pendulum is swinging back towards smart clients that can run AI locally, potentially reducing the need for massive data center infrastructure for inference.
Donald Thompson, Distinguished Engineer, LinkedIn