Feature embedding-based methods have performed exceptionally well in detecting industrial anomalies by comparing the features of the target image and the normal image. However, such approaches do not consider the inference speed, which is as important as accuracy in real-world applications. To relieve this issue, we propose a method called fast adaptive patch memory (FAPM) for real-time industrial anomaly detection. FAPM consists of patch-wise and layer-wise memory banks that save the embedding features of images in patch-level and layer-level, eliminating unnecessary repeated calculations. We also propose patch-wise adaptive coreset sampling for fast and accurate detection. FAPM performs well for both accuracy and speed compared to other state-of-the-art methods.