Augmented Reality (AR) appears as one of the most promising technologies for the next decades with a lot of applications in Industry 4.0. Unfortunately, most of existing systems are limited to small environments and running on a single device. To cope with these limitations, we present in this paper an AR Cloud solution for real-time crowd mapping. We propose an AR cloud service architecture that is able to 1) continuously relocalize AR devices in relation to the reference coordinate system of a common global 3D map in order to estimate the initial pose of the AR device and to correct tracking drift as soon as possible; 2) locally map in 3D the real environment surrounding each AR device; 3) fuse these 3D local maps with the global 3D map in order to keep it up-to-date, mandatory feature when real environment changes over time. Finally, we show engexperiments of our AR cloud platform at a large scale.