More simulation is not more data.

Synthetic English narration.

Keep the observed data fixed.
The observed moment comes from one finite data sample. Its sampling error remains fixed while we compare the model with that target.

Simulation adds another error.
At one candidate parameter, simulated data approximate the model's expected moment. With only a few simulations, that approximation contains simulation error.

Improve the model approximation.
Average more simulated data sets. Simulation error shrinks on average, although one particular sequence need not improve at every step. The observed target stays put.

Two different sources of error.
In this lesson, SMM approaches the direct GMM estimate for the same data sample. More simulation improves the model prediction; it does not create more observed data.