Background: High-resolution and longitudinal HIV molecular surveillance can inform the evolving hotspots to tailor regionally focused control strategies. Methods: HIV-1 pol sequences of three predominant genotypes (CRF01_AE, CRF07_BC, and CRF55_01B) were collected for molecular network reconstruction from people living with HIV (PLWH) in Guangzhou (2018–2020). They were categorized by geographical residences into central, suburban, and outer suburban areas. Clustering rates, assortativity coefficients, and intensity matrices were employed to assess transmission dynamics, geographic mixing patterns, and intra- and inter-area transmission, respectively. Results: Of the 2469 PLWH, 55.5% resided in the central area. Clustering rates showed no significant differences across areas (44.5%, 40.6% vs. 45.7%; p = 0.184). However, the transmission hotspots for CRF01_AE and CRF55_01B shifted to the outer suburban area. PLWH tended to form links within their local area (assortativity coefficient = 0.227, p < 0.001), particularly for CRF01_AE (0.512, p < 0.001; intra-area intensity = 69.2%). The central area exhibited the highest but decreasing intra-area transmission (74.5% to 30.2%), while intra- and inter-area transmission involving the outer suburban area increased (23.1% to 38.2%). Conclusions: Despite most PLWH residing in the central area, the outer suburban area emerged as the hotspot, requiring interventions towards both intra- and inter-area transmission.

Large HIV molecular clusters are often treated as homogeneous units, obscuring internal heterogeneity and the distinction between contemporary transmission hubs and inferred ancestral states. Using 935 CRF55_01B pol sequences (2008–2020, Guangzhou), we constructed a Tamura-Nei 93 molecular network. The 110-member giant component (GC) was partitioned via Louvain, and bridging members were identified. Phylogenetic analysis and PastML ancestral state reconstruction were performed, with sensitivity analysis at an alternative genetic distance threshold. Louvain identified 11 communities with a core-periphery architecture. Community 1 (C1) served as the dominant structural hub, driven by 21 bridging members who accounted for 82.5% of inter-community links and exhibited significantly higher degree centrality and shorter average shortest path length than non-bridging members (p < 0.05). However, ancestral state reconstruction most frequently assigned the root state to Community 2, revealing a striking discordance between structural centrality and inferred ancestry. At the 0.008 threshold, the GC expanded substantially, yet C1 persistently remained the primary hub, preserving this decoupling. Community-level analysis unravels hidden heterogeneity and decouples structural centrality from inferred ancestral state. Bridging-member identification highlights structurally important individuals for enhanced surveillance and further evaluation.