Energy storage such as battery and thermal energy storage is an effective approach to shift building peak load and alleviate grid stress at a building cluster level. However, due to the heterogeneous performance of different types of storage (e.g., response speed, charge/discharge efficiency and rate, storage capacity) and highly diversified energy use patterns of individual buildings, the multi-energy storage should be properly selected and optimally designed for individual buildings to achieve effective load shifting. The optimal deployment of multi-energy storage at a cluster level is a challenging optimization problem due to the nonlinear dynamic performance of the multi-energy storage and the high dimensionality as a result of a large number of buildings. To tackle the challenges, this study proposes a data-driven surrogate optimization method that optimally deploys multi-energy storage at a cluster level to minimize the building cluster energy bill under demand response programs. The method utilizes data-driven surrogate models to accurately predict demand response performance of individual buildings with multi-energy storage. An iterative optimization with automated energy-storage-option screening is developed to optimize the multi-energy storage configurations and design parameters. For a case study including 21 buildings, by optimally deploying multi-energy storage including battery, cooling TES tank, and building-integrated TES, the method reduced the building cluster energy bill by 8%–181% as compared to baseline cases. The optimal deployment method effectively identifies the buildings with better potential to adopt demand-side management and balances the pros and cons of the energy storage options, increasing demand response incentives by 12%–31%. The proposed method can be used in practice to facilitate the deployment of energy storage and improve engagement of buildings in demand response.