Machine learning, computer graphics and search engines all reduce to matrix arithmetic. Why does that convergence have practical consequences beyond being an elegant observation?

A. Because one optimised implementation serves all of them.

Effort spent optimising matrix multiplication benefits every field using it: GPUs were built to transform vertices, neural network training turned out to be chains of matrix multiplies, and the same hardware applied directly. It is why the standard advice is to express computation as matrix operations, not because it reads better, but because it routes work into optimised code.