Visual representation learning has made significant strides, yet existing approaches often overlook the crucial aspect of spatial understanding within images. While current self-supervised methods excel at learning semantic features, they typically do not explicitly encode location information, limiting their utility for downstream tasks that require spatial reasoning. We present a novel approach that learns location-aware visual representations through contrastive learning without requiring labeled data. Our method introduces an anchor-based strategy that explicitly encourages the model to capture both semantic and spatial information in its learned representations. By pioneering the integration of intra-image contrastive learning alongside traditional inter-image comparisons, our approach enables the extraction of rich spatial features while maintaining semantic understanding. We demonstrate the effectiveness of our learned representations through comprehensive similarity analysis, achieving 89.2% accuracy in spatial correspondence tasks - a 15-fold improvement over randomly initialized features. Our method opens new possibilities for self-supervised learning by showing how contrastive approaches can be extended to capture spatial relationships while maintaining semantic understanding.