The increasing demand for secure and confidential data processing in artificial intelligence (AI) models has led to the exploration of full homomorphic encryption (FHE). However, the practicality and efficiency of FHE in AI models remain unclear. This study investigates the impact of bit depth, number of CPU cores, and input text size on the computational speed and accuracy of FHE-based language models. The research aims to provide insights into the feasibility and limitations of FHE for large-scale AI applications. To achieve this goal, experiments were conducted using a range of encryption methods and parameters, with results showing significant slowdowns in execution time as bit depth increased. Additionally, CPU core count and input text size were found to have quadratic dependencies on generation time. The findings suggest that current solutions are too slow for mass production and are limited by quantization errors. The study's implications highlight the need for further research and development in FHE to improve computational resources, reduce algorithm complexity, and enhance encryption speed. By addressing these challenges, FHE may become a viable option for secure data confidentiality in AI models.