Résumé : Grid computing is a well-known and mature model of distributed computing based on the idea of collaborative sharing of resources between multiple users. Grids are generally used for executing large and long running applications. These applications are set up and configured to work on specific kinds of hardware and software. Compared to grid computing, cloud computing is an emerging model based on virtualization technologies, which provides on-demand resource leases as transparent services to the user. Distributed applications can easily run inside clouds, in already configured hardware and software environments. Whereas grids allow users to run very large and time-consuming applications on specific and dedicated hardware, cloud computing allows users to run applications in order to get fast results with a minimum of hardware and software configuration. Our vision is that, from a resource point of view, cloud computing is more dynamic and flexible than grid computing. We study and implement mechanisms in order to provide more dynamicity in grids by automatically extending them with cloud resources. Our work is based on XtreemOS, an open source, Linux-based grid operating system and Nimbus, an open source cloud computing framework. In this document, we present the design and implementation of a system using a Nimbus cloud in order to automatically provide on-demand resource nodes for an XtreemOS grid.