<p>The spatial distribution and flow patterns of high-end talents have been demonstrated to exert a significant influence on the innovation and development of regions. Conventional gravity models, constrained by linear assumptions, are inadequate in capturing the intricate nonlinear relationships and long-range spatial dependencies that are intrinsic to talent migration. The present paper puts forward the quantum gravitational field network (QGFN), a hybrid model that integrates the parallel advantages of quantum computing with the topological modelling capabilities of graph neural networks with a view to predicting and simulating high-end talent flows. The model employs hybrid amplitude-angle encoding to map urban features into a quantum state space, constructs gravitational potential functions via parameterised quantum circuits, extracts spectral information from the graph Laplacian using quantum phase estimation to compute topological attractiveness, and simulates dynamic migration processes through a Hamiltonian time evolution operator. Experiments on three real-world datasets—including intercity talent mobility in China, population migration across U.S. metropolitan areas, and cross-border talent flows in the European Union—demonstrate that QGFN reduces mean absolute error by 14.3–15.1% compared to the best baseline methods, and achieves a 17.6% improvement in 12-step prediction tasks, confirming its effectiveness in spatiotemporal network forecasting. This study provides a simulation‑based validation of a hybrid quantum‑classical framework for talent flow prediction. Actual quantum advantage on physical hardware remains to be demonstrated in future work.</p> Graphical abstract <p></p>