Data analytics and AI constitute key drivers for business innovation and are hence indispensable for ensuring the competitiveness of enterprises in the digital age. Yet, they can only be leveraged when sufficient data management is carried out, which requires the design and development of suitable data platforms. In recent years, so-called data lakehouses found their way into practice, which promise to combine the benefits of data warehouses and data lakes. However, only little is known about the data and technology architectures that are utilized for these data platforms in practice, as well as the experiences that enterprises have made with them. To address this gap, we conducted four case studies on large-scale, real-world data lakehouse implementations from the industrial sector that are used for various kinds of analytics and AI applications. This paper presents our within-case and cross-case results, which provide insights on common architectural decisions, practical experiences and observed challenges. They outline directions for future research and can support enterprises in the design of suitable data platforms.