Microglia and perivascular macrophages, myeloid-origin resident immune cells in the human brain, play crucial roles in Alzheimer's disease (AD)1-4. However, the field lacks a unified taxonomy describing their heterogeneity and plasticity5. To address this, we applied single-cell profiling to two independent, demographically diverse cohorts. The first comprises 543,012 viable myeloid cells from 137 unique postmortem brain specimens, while the second consists of 289,493 myeloid nuclei from 1,470 donors. Collectively, they cover the human lifespan and varying degrees of AD neuropathology. We identify 13 transcriptionally distinct myeloid subtypes, including the "GPNMB" subtype that proliferates with AD. We distinguish two contrasting homeostatic microglial states in AD and with aging: the first ("FRMD4A") wanes over time, while the second ("PICALM") becomes more prevalent. By prioritizing AD-risk genes, including PTPRG, DPYD, and IL15, and placing them into a regulatory hierarchy, we identify common upstream transcriptional regulators, namely MITF and KLF12, that regulate the expression of AD-risk genes in the opposite directions. Through the construction of cell-to-cell interaction networks, we identify candidate ligand-receptor pairs, including APOE:SORL1 and APOE:TREM2, associated with AD progression. We show polygenic risk for AD predisposes and prioritize the GPNMB subtype as a therapeutic target of early intervention. Our findings delineate the relationship between distinct functional states of myeloid cells and their pathophysiological response to aging and AD, providing a significant step toward the mechanistic understanding of the roles of microglia in AD and the identification of novel therapeutics.