Male infertility affects 10–15% of couples globally, with azoospermia — complete absence of sperm — accounting for 15% of cases. Traditional diagnostic methods for azoospermia are subjective and variable. This study presents a novel, noninvasive, and accurate diagnostic method using surface-enhanced Raman spectroscopy (SERS) combined with machine learning to analyze seminal plasma exosomes. Semen samples from healthy controls ($n=32$) and azoospermic patients ($n=22$) were collected, and their exosomal SERS spectra were obtained. Machine learning algorithms were employed to distinguish between the SERS profiles of healthy and azoospermic samples, achieving an impressive sensitivity of 99.61% and a specificity of 99.58%, thereby highlighting significant spectral differences. This integrated SERS and machine learning approach offers a sensitive, label-free, and objective diagnostic tool for early detection and monitoring of azoospermia, potentially enhancing clinical outcomes and patient management.