The scalability of a cost-effective and non-intrusive method for performance monitoring of multi-split type air conditioners is experimentally demonstrated on systems with different nominal capacities and design features. The method relies on an artificial neural network (ANN) technique, which uses easily measurable parameters representative of the refrigerant cycle and manipulated control parameters as the input for predicting the real-time cooling capacity of the system. Two air conditioning units intended for training and testing of the ANN model, are tested while being operated with their native control system in a broad range of experimental conditions for verifying the proposed methodology. The model scalability relies on the similarity of the fundamental working principles of vapor compression air conditioners. It is confirmed that temperature measurements in selected locations of the refrigerant pipeline and the total power consumption represents a suitable set of inputs for ensuring high prediction accuracy and ANN model scalability with rated system specifications from the product catalogue. The results show that the proposed monitoring method achieves accurate and generalizable predictions of the cooling capacity while limiting the relative error to below 10%.