Reliable wind speed prediction is critical for wind power scheduling and medium-to-long-term management. However, current operational forecasting systems often suffer from coarse spatial resolution, resulting in significant biases and limited practical applicability. This study applies a global stretched-grid dynamical downscaling approach to achieve high-resolution (HighRes, reaching 12.5 km in the focus region) sub-seasonal to seasonal (S2S) surface wind speed forecasts. This approach is implemented based on the standard-resolution operational version of the IAP-CAS ensemble prediction system (StdRes, with ~1° grid resolution). The added value of downscaling is evaluated using observations from over 2,000 stations across China. Results indicate that HighRes significantly enhances the fidelity of surface wind simulations, particularly in complex terrain and under weak wind conditions. Specifically, HighRes effectively corrects the positive wind speed bias prevalent in the StdRes model, with reductions in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) observed at 80 and 76% of stations, respectively. Improvements are most pronounced in the Northwest region and across various high-altitude stations (>3,000 m), reflecting enhanced representation of topographic drag. However, these improvements are wind-speed dependent, showing distinct advantages primarily under low wind speeds (<3 m/s). Analysis of year-round wind speed anomalies demonstrates that dynamical downscaling significantly improves amplitude prediction accuracy; RMSE reductions are observed at 70–80% of stations during lead weeks 1–4, whereas improvements in phase prediction are limited to the first 2 weeks. Seasonally, while overall prediction skill is higher in winter, the relative improvement from downscaling is superior in summer, with 60% of stations showing improved Temporal Correlation Coefficients (TCC) even at lead week 3. Furthermore, analysis of 850 hPa winds suggests that the negative surface wind bias in HighRes is associated with weakened lower-tropospheric winds, which are both fundamentally linked to the more realistic orography in HighRes and highlight the need for more careful treatment of terrain and related parameterizations in future high-resolution downscaling for wind applications.