To achieve the 2050 global climate target, offshore wind will increase to meet the growing demand of the direct and indirect electrification (e.g. green hydrogen production for the hard-to-abate sector). To keep up with the rapid increase of offshore wind generation, the energy balancing challenges due to the intermittency nature of wind and the network congestion/capacity challenges resulting from structural network capacity planning latency are to be addressed with the system integration technology. In this paper, it is proposed that the hydrogen electrolysis plant be co-located with the wind farm, of which the power consumption is controlled to track the wind generation profile accurately to cancel the intermittency of wind generation, and reduce the required grid connection capacity, and thereby avoid the expensive grid expansion. However, this power tracking controller introduces a cross-plant feedback path from the wind farm to the hydrogen plant, posing challenges for system partitioning in stability analysis, which also makes it difficult to design the power tracking control with a good trade-off between the tracking performance and stability margins. To address this issue, this paper proposes an equivalent transformation to eliminate the cross-plant feedback path. Then, the dynamic power stability analysis with different partition methods is examined, which are mathematically proven to be equivalent in terms of stability conditions, but provide different insights. An optimal partition method is then proposed in this paper, which not only provides clear insight on the ideal and non-ideal power tracking performances but also can also identify the stability issues of different minor loops individually. Finally, the proposed optimal partition method and its valuable insights into power tracking performance and stability analysis are validated through timedomain simulations of a 180 MW integrated wind-to-hydrogen plant with a realistic complexity.