Seawater recirculating aquaculture is a sustainable aquaculture method that provides benefits such as the conservation of water and land resources, high productivity, and environmental protection. The high NO3–-N concentration in wastewater is a primary factor limiting wastewater discharge to meet compliance standards and restricting the sustainable development of seawater recirculation aquaculture. Biological denitrification is the primary method for removing NO3–-N in the water. Sulfur autotrophic denitrification (SAD) does not require an external organic carbon source and produces low sludge production, making it suitable for treating seawater recirculating aquaculture water with lower C/N ratios. However, SAD generates H+, which reduces the pH of water, thus affecting the stability of the denitrification device in long-term operation. In the actual operation of the SAD device, oyster shells are frequently used as a filler substrate to regulate the pH of water and ensure the effectiveness of the device in denitrification. Oyster shells, as kitchen waste, are cheap and easy to obtain and have received widespread attention. Studies on the denitrification performance and microbial community structure of the SAD device for marine recirculating aquaculture wastewater with varying hydraulic loading rates (HLRs) and S0/Oyster shell filling ratios are limited. In this study, we compared the denitrification performance of SAD devices with three S0/Oyster shell ratios (5:1, 3:1, and 1:1) under five HLRs [0.19, 0.24, 0.32, 0.48, and 0.95 m3/(m2·d)] and the changes in influent and effluent pH and DO in the treatment of seawater recirculating aquaculture wastewater, using artificial seawater recirculating aquaculture wastewater as the treatment target. The effects of different S0/Oyster shell ratios on the nitrogen removal performance of the SAD device were evaluated in combination with microbial community characterization and functional gene prediction analysis. When the HLR was 0.19~0.48 m3/(m2·d), no significant difference was observed in the NO3–-N removal rates among the four HLRs and three devices, which were (72.11±12.64)%~ (75.85±7.95)%, (76.00±6.91)%~(78.13±6.45)%, (70.40±7.78)%~(75.76±8.98)%, respectively. At the highest HLR [0.95 m3/(m2·d)], the NO3–-N removal efficiency of the three devices significantly decreased, and the NO3–-N removal efficiency of the S0/Oyster shell=5:1 (61.16%±9.31%) and 3:1 (56.62%±7.23%) devices was significantly higher than that of the S0/Oyster shell=1:1 (38.98%±10.19%). For the S0/Oyster shell=3:1, the average concentration of effluent NO2–-N of the device was the lowest at (0.59±0.39) mg/L. No significant difference was observed in the average concentration of effluent NH4+-N among the three devices, ranging from (0.17±0.07) to (0.19±0.11) mg/L. The denitrification performance of S0/Oyster shell=5:1 and 3:1 devices was better. The effluent pH of the device decreased with increased S0/Oyster shell ratio and HLR. The dominant bacterial phyla in the SAD device were Campilobacterota (6.47%~59.73%) and Proteobacteria (16.46%~53.93%), and the dominant bacterial genus was Sulfurimonas (2.70%~49.50%). As the ratio of S0/Oyster shells decreased, the abundance of Sulfurimonas increased within the device and at the intersection of oyster shells and S0 in the upper part of the device. pH was positively correlated with denitrification gene abundance. This study provides basic theoretical data for the design and operation of SAD devices in seawater RAS.