With the widespread application of multi-view data in real-world scenarios, addressing clustering problems with incomplete views has become a significant challenge in machine learning. Traditional anchor graph clustering methods rely on complete data to build anchor graphs but face two main issues: limited anchor representation under high missing rates and not fully using the strengths of different views under low missing rates. To address three key limitations—restricted anchor selection, fixed weight assignments, and high computational costs, this paper proposes the following solutions: First, a cross-view anchor complementarity mechanism selects anchors from both common and unique samples across views, improving representation accuracy under severe missing conditions. Second, a missing-pattern-aware weighting model automatically adjusts each view’s contribution to the similarity matrix according to sample missing patterns. Third, this paper employs doubly stochastic non-negative matrix factorization to optimize spectral clustering complexity, reducing it from cubic to linear time complexity relative to sample size. Based on these innovations, this paper develops the Incomplete Multi-View Clustering with Sample-Adaptive Complements (IMVC-SAC). Experiments on five standard datasets show that IMVC-SAC outperforms existing methods, especially maintaining strong performance when 70%-90% of data is missing, proving its effectiveness in real-world scenarios.