<p>Internal pore defects in metallic materials significantly degrade fatigue performance, leading to life scatter and uncertainty. Quantifying this uncertainty can enhance the accuracy of fatigue life predictions, which is of vital for elucidating the influence mechanisms of critical manufacturing defects. This study develops a physics-data co-driven methodological framework to predict the fatigue life distribution of metallic materials with multiple pore defects based on the <i>M</i>-integral fatigue model, physics-informed neural network (PINN), and Gaussian process regression (GPR). Cyclic loads (Δσ) and the equivalent damage area (<i>A</i>_<i>D</i>) are employed to estimate the lifetime uncertainty arising from different defect characteristics. In the proposed method, <i>A</i>_<i>D</i>—derived from an <i>M</i>-integral equivalent calculation—quantifies multiple pore defects. A physical loss function based on the <i>M</i>-integral fatigue model enhances the accuracy of the PINN and reduces its dependency on experimental data. The lifetimes predicted by the PINN serve as inputs to the GPR, which subsequently generates a probability density function, yielding the prediction interval for fatigue life. Numerical results demonstrate that the proposed model can accurately predict the fatigue life distribution (95% confidence interval) of materials with multiple pore defects using limited training data, achieving a coefficient of determination (<i>R</i>²) above 0.8. This work addresses the existing gap in methods for predicting the fatigue life uncertainty of materials containing multiple pores, providing a robust methodology for the quantitative analysis of multi-defect characteristics.</p>