A grid coordinator which has been configured for autonomic failover can include a monitor communicatively linked to a multiplicity of grid hosts in a services grid. A metrics store can be coupled to the monitor and configured to store service metrics for individual service instances in the grid hosts. Optimization logic can be programmed to compute a best-fit between metrics stored for a set of service instances in a failed grid host, and platform metrics determined for a proposed replacement grid host. Finally, a failover processor can be coupled to the monitor and the optimization logic and communicatively linked to the grid hosts to create a new set of service instances in the proposed replacement grid host to replace the set of service instances in the failed grid host according to the best-fit in the optimization logic.