Climate warming poses severe threats to soil organic carbon (SOC) stocks, particularly in cryospheric regions exemplified by the Qinghai-Xizang Plateau (QXP). The divergent responses of particulate organic carbon (POC) and mineral-associated organic carbon (MAOC) to global warming necessitate precise pool-specific quantification for robust climate solutions. Here, we sampled soils from 220 field sites and combined with Markov Chain Monte Carlo data assimilation (MCMC) and deep learning modeling, generated high-resolution (1 km × 1 km) POC and MAOC stocks maps in QXP. QXP stores 9.8 Pg POC and 13 Pg MAOC in 0–40 cm soils, with spatial trends primarily governed by temperature gradients. MAOC accumulation reflects C input by vegetation and POC stocks are controlled by microbial decomposition. Warming decreased POC and MAOC in low-temperature regions when ground surface temperature below + 5 ℃ and mean annual temperature under + 3 ℃, respectively. Above this threshold, POC and MAOC increased for enough plant-C inputs and microbial activities. Global warming until 2060 s generate escalating C loss risks, from −1.0 % in SSP1-1.9 to −2.6 % in SSP5-8.5, particularly due to low POC stability and amplified microbial activity. This study advances pool-specific C accounting in cryospheric high elevation ecosystems while quantifying climate vulnerability thresholds critical for adaptive management. © 2025 Elsevier B.V., All rights reserved.