Techniques for enabling end-to-end compression when performing host-based replication (HBR) are provided. In one set of embodiments, a hypervisor running on a host system at a primary site can intercept I/O write requests that originate from a virtual machine (VM) configured to be replicated via HBR, the I/O write requests being destined for a virtual disk (VMDK) of the VM. The hypervisor can further track VMDK file blocks that are modified by the I/O write requests and can retrieve the VMDK file blocks from a storage tier at the primary site. The hypervisor can then compress the retrieved VMDK file blocks and transmit the compressed blocks to a secondary site.