High-definition (HD) semantic maps are crucial in enabling autonomous vehicles to navigate urban environments. The traditional method of creating offline HD maps involves labor-intensive manual annotation processes, which are not only costly but also insufficient for timely updates. Recent studies have proposed an alternative approach that generates local maps using online sensor observations. However, this approach is limited by the sensor's perception range and its susceptibility to occlusions. In this study, we propose Neural Map Prior (NMP), a neural representation of global maps. This representation automatically updates itself and improves the performance of local map inference. Specifically, we utilize two approaches to achieve this. Firstly, to integrate a strong map prior into local map inference, we apply cross-attention, a mechanism that dynamically identifies correlations between current and prior features. Secondly, to update the global neural map prior, we utilize a learning-based fusion module that guides the network in fusing features from previous traversals. Our experimental results, based on the nuScenes dataset, demonstrate that our framework is highly compatible with various map segmentation and detection architectures. It significantly improves map prediction performance, even in challenging weather conditions and situations with a longer perception range. To the best of our knowledge, this is the first learning-based system for creating a global map prior.