Increase number of artificial intelligence (AI) and high-performance computing (HPC) applications call for advanced hardware. However, after decades of innovation, Moore’s law seems to have hit a plateau. One of the potential avenues to keep the hardware technology advancing is to heterogeneously integrate multiple small chips (chiplets) at the package level. E.g. 2.5D microelectronic packages with silicon interposer integrating multiple chips have already entered high volume manufacturing. Some of the key challenges for such packages are low assembly yield due to high package warpage, high cost due to large silicon footprint. To overcome these challenges, the high-density organic substrates with attached fine-pitch redistribution layers (RDL) have been developed targeting line widths and pitch of 2 µm or less. In such applications, the substrate consists of the high-density circuitry layers for die-to-die interconnections attached to conventional base carrier comprised of organic materials for better power delivery network and high speed data transmission. Although, such organic substrates with built-in fine pitch interposer are cheaper, their assembly yield is low. One of the reasons for low assembly yield is high incoming bare substrate warpage. In this study, we present an AI driven methodology to reduce the warpage of such asymmetric substrates by optimizing the copper volume percentage on each backside layer of the core. We start by creating a finite-element model to accurately predict the warpage. Next, a radial basis function (RBF) based AI surrogate for the finite-element model is trained to predict the substrate warpage. Genetic algorithm is then used to minimize the warpage predicted by the surrogate by varying the copper percentage distribution in the backside layers of the base carrier. The optimized volume percentages are then input back into the finite-element model to validate the optimization results. It is seen that such an approach can be used to reduce the warpage by 70% compared to the baseline.