Introduction: Complete surgical eradication of cholesteatoma is essential to prevent recurrence, but intra-operative differentiation from inflammatory mucosa and other middle ear structures remains a major surgical challenge. Under standard white-light microscopy, residual microscopic disease is frequently missed, leading to high recurrence rates, especially in Canal-Wall-Up surgeries. Hyperspectral imaging (HSI) captures unique spectral signatures of tissues beyond the visible light spectrum. We hypothesise that HSI, when coupled with an artificial intelligence (AI) model, can provide objective, real-time, intraoperative identification of cholesteatoma and localising areas at high risk for recurrence. Methods: Ex-vivo surgical specimens of cholesteatoma and four other middle-ear tissue types were collected during routine otological surgery. Each specimen was imaged using a hyperspectral camera that covers the visible to near-infrared range. An experienced otologist determined ground truth. A machine learning model was trained on the labelled spectral–spatial data to learn the unique spectral “fingerprint” of cholesteatoma. Results: A preliminary dataset of 233 images from 17 patients, encompassing cholesteatoma and a range of non-cholesteatoma tissue were used for training and evaluation. Our model differentiated cholesteatoma from non-cholesteatoma tissue with an overall pixel-wise accuracy (OA) of 72.53%, an average accuracy (AA) of 69.14%, a sensitivity of 73.40%, and a specificity of 76.62%. Conclusion: An AI-driven HSI system can accurately differentiate cholesteatoma from normal middle-ear tissue in a surgical setting. This label-free, non-ionising modality shows promise as a real-time surgical guide. Patient recruitment and imaging are ongoing, and based on observed learning-curve trends, we anticipate substantial performance gains as the hyperspectral sample size increases. Its integration into the operative workflow may improve the completeness of disease removal and reduce recurrence.