The rapid expansion of Artificial Intelligence (AI) has led to a significant increase in the use of Data Centres (DCs), which are essential for processing and storing vast amounts of data. However, this surge in AI deployment has raised environmental concerns about increased Carbon Dioxide (CO2) emissions. Various solutions have been proposed to address the energy efficiency of DCs such as advanced cooling systems or selecting training locations with lower cooling needs or greener power supplies. To achieve further improvements, one needs to be able to measure actual emissions at the code level so that an optimization strategy can be designed and evaluated. To address the issue, we explore an innovative approach to precisely measure the CO2 emissions of AI applications. By introducing a linear regression energy estimation model based on Performance Monitoring Counters (PMCs) we calculate the CO2 emission of AI applications. PMCs such as the total number of instructions and the total number of cycles of the computer processor are considered ideal for energy estimation due to their strong correlation with the processor’s energy consumption and minimal overhead on resource utilisation. For this research, only the Central Processing Unit (CPU) and Dynamic Random Access Memory (DRAM) are considered, as they consume the maximum energy compared to other parts of the processor. This approach is easily extendable to GPUs. In the presented evaluation, the energy estimation model produced an error of only 0.158% for CPU and 0.272% for DRAM.