A method and an apparatus for managing resources of a network slice are disclosed. In the method, resources are allocated more properly by using a feature that peaks and valleys of load of network slices overlap, so that resources can be coordinated from another network slice whose independent decision result is scale-in or is keeping a quantity of instances unchanged to a network slice whose independent decision result is scale-out, or resources can be coordinated from another network slice whose independent decision result is scale-in to a network slice whose independent decision result is keeping a quantity of instances unchanged, and therefore a network slice whose capacity needs to be increased does not need to request new resources from an infrastructure, thereby improving resource utilization of the infrastructure and reducing resource procurement costs of a slice tenant.