Compute Express Link (CXL) has emerged as a frontier for disaggregation by providing a fabric supporting memory accessing, caching, and peer-to-peer memory access between devices. CXL externalizes the internal memory fabric of a server and blurs the notion of server for realistic disaggregation. The key feature of enabling of CXL as a memory fabric is its support of multi-level switching up to 4096 endpoints. However, CXL's current multi-level switching poses challenges on scalability and latency. To cope with the expected scale of CXL fabric and take full advantage of disaggregation, we propose Aurelia. Aurelia architects addressing, routing, and transport as networking layers, which are typical on host networking, for CXL fabric. Aurelia uses existing CXL mechanism to realize the much-needed functionalities for a scalable fabric. With these networking layers, Aurelia will be able to support future disaggregation workloads at scale.