Soil could represent a potentially notable source of carbon for achieving global carbon neutrality. However, how the land surface soil organic carbon (SOC) stock, which is more sensitive to climate change than other carbon stocks, will change naturally under the influence of global warming remains unknown. In this work, the global land surface SOC trends from 1981–2019 were explored, and the driving factors were identified. A random forest model (a type of machine learning method) was proposed to predict future global surface SOC trends integrated with climate scenarios of the Coupled Model Intercomparison Project Phase 6 (CMIP6) models. The results revealed that the global surface SOC content will increase, while the temperature and precipitation are the main climate drivers at the global scale, and vegetation cover is a crucial local factor influencing the increase in SOC. However, under the 1.5 °C global warming scenario, the land SOC sink will increase by 13.0 petagram carbon (PgC) at most compared with that under the SSP2-4.5 scenario, which accounts for only 19% of the total carbon emission capacity at the current 1.1 to 1.5 °C global warming level. Moreover, this value is far from the Paris Agreement target of four out of one thousand for the annual increase in the soil carbon stock 40 cm below the surface over the next 20 years (2.72 PgC·a⁻¹). This illustrates that overreliance on natural carbon sinks is a high-risk strategy. These findings highlight the urgency of implementing mitigation and removal strategies to reduce greenhouse gas emissions.