Despite more than two-thousandfold variation in genome size, key features of genome architecture are largely conserved across flowering plants. Parasitic plants have elucidated the many ways in which genomes can be modified, yet we still lack comprehensive genome data for species that represent the most extreme form of plant parasitism. Here, we report the genome of the endophytic parasitic angiosperm Sapria himalayana Griff. (Rafflesiaceae), which lacks a typical plant body and challenges several fundamental assumptions of plant genome conservation. 44% of the genes conserved in eurosids are lost in Sapria, dwarfing any previously reported level of gene loss in vascular plants. These losses demonstrate remarkable functional convergence when compared to other parasitic plants, suggesting a common genetic roadmap underlying the evolution of plant parasitism. Meanwhile, at least 1.2% of the Sapria genome, including both genic and intergenic content, is inferred to be derived from host-to-parasite horizontal gene transfers (HGTs) and includes genes potentially adaptive for parasitism. Focused phylogenomic reconstruction of these HGTs reveal a hidden history of former host-parasite associations involving close relatives of their modern hosts in the grapevine family. Our findings offer a unique perspective into how deeply angiosperm genomes can be altered to fit an extreme form of plant parasitism. These results also demonstrate the utility of HGTs as DNA fossils to investigate extinct symbioses.