Deadbeat predictive current control (DPCC) exhibits high control dynamics hence is very attractive for high-speed permanent magnet synchronous machine (PMSM) drives. However, the control performance of DPCC heavily relies on machine parameters and would be significantly deteriorated if inaccurate machine parameters are employed. The parameter robustness of DPCC would further deteriorates at high speeds as the switching-to-fundamental frequency ratio (SFR) decreases. Therefore, to improve the robustness of DPCC for high-speed PMSM drives, this paper introduces a deadbeat prediction method based on the incremental model and extended state observer (ESO), which does not require the magnetic linkage parameters and exhibits excellent control performance under high-speed condition. The proposed method is compared to conventional DPCC and conventional DPCC based on incremental model, and the effectiveness has been confirmed by extensive simulations.