The cooperative simultaneous localization and mapping problem has acquired growing attention over the years. Even though mapping of very large environments is theoretically quicker than a single robot simultaneous localization and mapping, it has several additional challenges such as the map alignment and the merging processes, network latency, administering various coordinate systems and assuring synchronized and updated data from all robots and also it demands massive computation. This article proposes an efficient architecture for cloud-based cooperative simultaneous localization and mapping to parallelize its complex steps via the multiprocessor (computing nodes) and free the robots from all of the computation efforts. Furthermore, this work improves the map alignment part using hybrid combination strategies, random sample consensus, and inter-robot observations to exploit fully their advantages. The results show that the proposed approach increases mapping performance with less response time.