The brain runs on 20W. It doesn't do matrix math. It routes sparse signals through a graph where each node makes a local weighted decision. Intelligence emerges from the path, not from a global forward pass.

GPUs were adopted for AI because matrix multiplication is embarrassingly parallel. But is matrix multiplication actually required for intelligence — or is it just the only method we've tried at scale?