This study compared the efficacy, safety, and intracytoplasmic sperm injection (ICSI) outcomes of artificial intelligence (AI)-assisted versus conventional microdissection testicular sperm extraction (micro-TESE) in men with non-obstructive azoospermia (NOA). To this end, we retrospectively analyzed 1286 consecutive NOA patients undergoing micro-TESE at the two reproductive medicine center of Sichuan Provincial People's Hospital (Chengdu, China) and Sichuan Jinxin Xinan Women's and Children's Hospital (Chengdu, China) from January 2020 to December 2024, including 542 patients receiving AI-assisted procedures and 744 undergoing conventional surgery. The primary outcome was sperm retrieval rate (SRR), while secondary outcomes covered intraoperative parameters, postoperative complications, and ICSI outcomes, and univariable and multivariable logistic regression were performed to identify independent predictors of SRR and live birth. The results showed that the overall SRR reached 56.2% in the entire cohort, and the AI group had a significantly higher SRR (68.1% vs 47.6%, P < 0.001), with the most prominent improvement observed in patients with Sertoli cell-only syndrome (47.2% vs 22.7%, P < 0.001). The AI group also exhibited shorter operative time, less intraoperative blood loss, fewer biopsy sites, shorter postoperative hospital stay, and lower complication rates (all P < 0.001), as well as superior ICSI outcomes including higher fertilization, high-quality embryo, blastocyst, clinical pregnancy, ongoing pregnancy, and live birth rates (all P < 0.01). Further multivariable analysis confirmed that AI-assisted surgery serves as an independent predictor of both successful SRR (odds ratio [OR] = 3.182, 95% confidence interval [CI]: 2.374-4.265, P < 0.001) and live birth rate (OR = 1.876, 95% CI: 1.325-2.656, P < 0.001). Taken together, AI-assisted micro-TESE improves sperm retrieval efficiency, reduces perioperative morbidity, and enhances ICSI and live birth outcomes in NOA patients as an independent predictor for both SRR and live birth rate, and prospective multicenter randomized trials are warranted for further validation.