5G service providers consider Network Function Virtualization (NFV) as an enabler to foster new opportunities to scale their business while reducing operational expenses. NFV builds on cloud native technologies, automation, and reusability. Central to the success of NFV is the ability to design service templates once and to on-demand deploy those template services into specific service contexts, e.g., network slices. Therefore, on-demand homing and assigning of service designs with service level requirements over distributed cloud resources and capabilities is one of the main challenges for the transition to NFV. Towards this end, it is important to embed service graphs over the given infrastructure such that the total cost is minimized while satisfying service requirements. To get around the lack of scalability of optimization-based approaches, a viable approach is to rely on efficient Virtual Network Function (VNF) embedding heuristics. Despite their low complexity, VNF embedding heuristics suffer from the so-called causality issue, which means that embedding decisions must be made before all neighboring dependencies were embedded. This, as a result, may lead to an inferior embedding outcome. In this paper, we propose our novel h-horizon sequential greedy look-ahead embedding algorithm, which allows embedding and re-embedding of VNFs based on embedding decisions of other VNFs to overcome the causality issue. Our simulation results indicate that our proposed algorithm outperforms the existing greedy benchmark in terms of total embedding cost.