Marine biogeochemical change unfolds across multiple timescales, yet how ecological risk reorganizes in human-pressured marginal seas remains poorly constrained. Here we use an upper-ocean biogeochemical reconstruction of the China Seas (1991–2020, 0.25°×0.25°) to quantify threshold-defined heat, low-oxygen, chlorophyll-anomaly and nutrient-enrichment extremes and their annual co-occurrence as a compound ecological index (CEI). We show that ecological risk in the China Seas is shifting disproportionately towards compound exposure in past 30 years. Single-risk exposure increased by approximately 45%, whereas CEI≥2 compound-risk exposure increased more than threefold, nearly doubling the share of compound risk within all risk-exposed waters. This transition was expressed as an intensification of nearshore compound-risk cores and an expansion towards inner- and middle-shelf environments, with the middle shelf showing the fastest relative increase from a low baseline. The changing risk structure was strongly type-specific: nutrient-enrichment risk showed the largest single-risk expansion, while chlorophyll-anomaly–nutrient-enrichment combinations and low-oxygen–nutrient-enrichment combinations dominated late-period compound risk. Risk amplification also followed a “bad-to-worse” pattern, whereby pre-existing hotspots were more likely to intensify or upgrade into higher-order compound states. Ecologically, fish habitats became increasingly exposed to compound conditions, with compound-risk exposure within spawning grounds rising from 10% to 30%, concentrated mainly nearshore and on the inner shelf. Diagnostic pathway analysis indicates that resolved compound-risk expansion was most broadly consistent with riverine nutrient–productivity pressure, while climate–ventilation pressure and dual-pathway coupling were also important. These pathway-consistency results reveal a modern transition of marginal-sea biogeochemistry from isolated hazards towards compound ecological exposure, providing constraints on future marginal-sea sensitivity, resilience and ecological-risk thresholds.