Brain age has emerged as a crucial biomarker in neuroscience research, where MRI-based methods currently dominate due to their established predictive accuracy. However, MRI’s prohibitive cost and technical constraints severely limit its clinical scalability. While EEG offers a cost-effective alternative, existing EEG-driven approaches have consistently underperformed in brain age estimation, with reported MAE values typically exceeding 6 years in prior studies. To overcome these limitations, we present BiGDC-BrainAgeNet – the first graph neural architecture specifically designed for EEG-based brain age estimation. Our model synergistically integrates a bidirectional graph diffusion convolutional gated recurrent unit with self-attention mechanisms, enabling simultaneous capture of spatiotemporal dynamics and global feature interdependencies within EEG signals. This innovative framework achieves state-of-the-art performance with unprecedented accuracy, delivering MAEs of 4.026 years ( \(R^2\)  = 0.83) and 1.756 years ( \(R^2\)  = 0.885) on the TUAB and CHBMP datasets respectively. Spatial analysis further reveals that central sulcus and occipital electrodes contribute most significantly to age prediction, providing novel neurophysiological insights. As the first successful application of graph neural networks in EEG-based brain aging research, our work establishes a new framework for affordable, large-scale brain health monitoring.