(1) Background: While the antidepressant effects of aerobic exercise (AE) are well documented, fewer studies have examined impact of AE as an add-on treatment. Moreover, various effects on neurobiological variables have been suggested. This study examines effects of AE on Cortisol Awakening Reaction (CAR), serum Brain Derived Neurotrophic Factor (sBDNF), Tumor Necrosis Factor alpha (TNF-alpha) and sleep. (2) Methods: Inpatients with moderate-to-severe depression (N = 43) were randomly assigned to the AE or stretching condition (active control) taking place 3x/week for 6 weeks. CAR, sBDNF and TNF-alpha were assessed at baseline, after 2 weeks and post-intervention. The 17-item Hamilton Depression Rating Scale (HDRS17), subjective sleep quality measured by the Pittsburgh Sleep Quality Index (PSQI) and polysomnography (PSG) were obtained at baseline and post-intervention. (3) Results: Stress axis activity decreased in both groups from baseline to post-intervention. sBDNF showed a significant increase over time, whereas the number of awakenings significantly decreased. No significant time by group interactions were detected for any of the study variables. Correlational analyses showed that higher improvements in maximum oxygen capacity (VO2max) from baseline to post-intervention were associated with reduced scores on the HDRS17, PSQI and REM-latency post-intervention. (4) Conclusions: While some neurobiological variables improved during inpatient treatment (CAR, sBDNF), no evidence was found for differential effects between AE and an active control condition (stretching). However, patients in which cardiorespiratory fitness increased showed higher improvements in depression severity and depression-related sleep-parameters.