The usual approach to this kind of question is not to ask how many latent variables you have, but rather how many parameters are in your model. For example, if you have 50 observed variables and you allow all your latent variables to be correlated, you have 50 loadings and 45 correlations = 95 parameters to be estimated.
From there, things get less clear because there are a variety of recommendations for how many observations you should have per parameter, with either 5 or 10 per being common suggestions.