“What we're seeing is all kinds of demand for different architectures and compute type of solutions to run inference not in ...