Cross-locale product recommendation presents unique challenges, particularly in underrepresented locales where data availability is limited. This study explores the application of large language models (LLMs), including ChatGPT4o-Mini, Gemini-1.0-Pro, and Gemini-1.5-Flash, for next-item prediction on the Amazon-M2 dataset. Using a combination of fine-tuning and prompt engineering, this work demonstrates that LLMs outperform traditional baseline models and state-of-the-art session-based methods such as GRU4Rec++, achieving significant improvements in MRR@100 and Recall@100. Furthermore, the study evaluates the use of user-preference summaries as an alternative to directly listing previous product titles to optimize token usage and computational efficiency. The findings highlight the potential of LLMs for accurate and efficient cross-locale product recommendations while outlining future directions to improve scalability and efficiency.