Reanalysis data and a numerical model are employed to explore how diffusion in the upper ocean evolves in different seasons and to understand its contribution to seasonal heat storage (SHS) and Arctic amplification. The numerical simulation results are closely consistent with observations. First, Arctic solar radiation absorption anomaly (maximum in June) due to ice‐loss is mainly stored as SHS (maximum in June and July) or warms the surface (minimum in July) in observations. Furthermore, the numerical simulation suggests that vertical diffusion dominates SHS formation/discharge. Second, surface ocean becomes warmer than surface air in cold season and releases SHS in Arctic. Sea‐ice loss allows more ocean be directly driven by wind; surface high pressure over Arctic may enhance the wind stress; increased water freeze leads to more salt rejection and additional dense water being produced and sinking downward; all of the occurrences strengthen vertical mixing and release more SHS to the atmosphere. As a result, Arctic surface warming reaches its maximum in cold season. In observations, much stronger outgoing longwave radiation due to the warmer surface in October causes Arctic surface warming to reach its maximum in November, although increased carbon dioxide forcing also contributes to maintain longwave radiation in November. Finally, Arctic surface (air) warming in cold season exhibits different spatial patterns from SHS discharge due to heat convergence/divergence induced by anomalous surface wind, which is determined by the variation in surface high pressure over the Arctic.