Metal sintering is a necessary step for the fabrication of metal-injection-molded parts and binder jetting techniques such as that in HP’s metal 3D printer (MetJet). This process induces significant deformations, typically ranging from 25 to 50%, depending on the porosity of the green part (printed part before post-processing). Achieving precise geometrical accuracy and consistency in the final part presents a substantial challenge to increasing the manufacturing yield. This challenge is primarily attributed to the high porosity of green parts produced by MetJet (compared with alternative technologies such as metal injection molding), which can lead to approximately 50% volumetric shrinkage after sintering. Moreover, this shrinkage is nonisotropic depending on nonuniform stress built up during sintering, resulting in deformations such as gravitational sag, gravitational slump, and surface drag. In this study, we employ a graph-based deep learning approach to predict the deformation of the part where deformation simulation can be substantially sped up at the voxel level. By utilizing a well-trained metal sintering inferencing engine, the final sintering deformation value can be obtained in a matter of seconds. The tested accuracy on a sample complex geometry achieves a mean deviation of 0.7 um for a 63 mm test part in a single sintering step (equivalent to 8.3 min of physical sintering time) and a mean deviation of 0.3 mm for the complete sintering cycle (~4 h of physical sintering time).