They say linear in time for increasing circuit depth not increasing qubits. They also don’t address the classical algorithm being exponential in memory as well.