Permafrost degradation on the Tibetan Plateau is well-documented and expected to continue throughout this century. However, the impact of thawing permafrost on the distribution, composition, and resilience of vegetation communities in this region is not well understood. In this study, we combined a transient numerical permafrost model with machine learning algorithms to project the near-future thermal state of permafrost and vegetation (represented by the Normalized Difference Vegetation Index [NDVI]) changes under two contrasting climate pathways (Shared Socioeconomic Pathway 1–2.6 [SSP1–2.6] and SSP5–8.5). The contribution of climatic and terrestrial variables to vegetation evolution was quantified using ridge regression. By 2100, permafrost areas were expected to decrease by 21±4%, and 55±2% under the SSP1–2.6 and SSP5–8.5 scenarios, respectively, relative to the baseline period (2000–2018). Under the SSP1–2.6 scenarios, the mean annual ground temperature and active layer thickness were projected to fluctuate stably, while under the SSP5–8.5 scenarios, a significant increasing trend was anticipated. Satellite-based observations indicated an increasing trend of NDVI within the permafrost areas from 2000 to 2018 (0.01 per decade), mainly attributed to climatic factors. In the future, vegetation greenness was expected to possibly remain stable under SSP1–2.6 scenarios, whereas a rising trend was likely noted under SSP5–8.5 scenarios during 2019–2050, mainly controlled by the surface air temperature and liquid water content at the root zone during the growing season. Our modeling work provides a potential approach for investigating future vegetation changes and offers more possibilities to improve understanding of the interaction between soil-vegetation-atmosphere in cold regions.