Mobile Edge Computing: A Comprehensive Review of LLM Inference Offloading, Resource Optimization, and Open Research Challenges

Mobile Edge Computing offers a practical foundation for bringing LLM intelligence closer to users, but efficient deployment requires more than simply moving inference from the cloud to an edge server. LLM inference is characterized by large model footprints, autoregressive dependencies, dynamic KV-cache requirements, heterogeneous hardware, and sensitivity to network conditions. Consequently, inference location, model configuration, communication resources, and compute resources should be optimized jointly. This review has examined local inference, full and partial offloading, model partitioning, collaborative execution, quantization, pruning, distillation, speculative decoding, KV-cache optimization, caching, learning-based orchestration, and mobility-aware resource management. The most important future opportunity is the development of adaptive systems that treat the model, device, network, and edge infrastructure as one optimization space. Such systems can provide low latency and energy consumption while preserving model quality, privacy, and reliability. As 5G-Advanced and 6G networks mature, MEC-based LLM inference is positioned to become a key enabling technology for responsive, privacy-aware, and resource-efficient edge intelligence.

1.      Chen, Z., Zhu, B., Wang, J., Shin, H., Nallanathan, A., & Niyato, D. (2026). Network Edge Inference for Large Language Models: Principles, Techniques, and Opportunities. ACM Computing Surveys, 58(12). DOI: 10.1145/3809166. 2.      Cai, G., Tian, R., Yang, L., Jia, Y., Li, L., & Wang, J. (2026). Efficient Inference for Edge Large Language Models: A Survey. Tsinghua Science and Technology, 31(3), 1365–1380. DOI: 10.26599/TST.2025.9010166. 3.      Wang, R., Gao, Z., Zhang, L., Yue, S., & Gao, Z. (2025). Empowering large language models to edge intelligence: A survey of edge efficient LLMs and techniques. Computer Science Review, 57, 100755. DOI: 10.1016/j.cosrev.2025.100755. 4.      Dong, S., Tang, J., Abbas, K., Hou, R., Kamruzzaman, J., Rutkowski, L., & Buyya, R. (2024). Task offloading strategies for mobile edge computing: A survey. Computer Networks, 253, 110791. DOI: 10.1016/j.comnet.2024.110791. 5.      Shakarami, A., Ghobaei-Arani, M., & Shahidinejad, A. (2020). A survey on the computation offloading approaches in mobile edge computing: A machine learning-based perspective. Computer Networks, 182, 107496. DOI: 10.1016/j.comnet.2020.107496. 6.      Qu, G., Chen, Q., Wei, W., et al. (2025). Mobile Edge Intelligence for Large Language Models: A Contemporary Survey. IEEE Communications Surveys & Tutorials. DOI: 10.36227/techrxiv.172115025.57884352. 7.      Zhou, X., & Liu, J. (2025). DAPO: Mobility-Aware Joint Optimization of Model Partitioning and Task Offloading for Edge LLM Inference. Electronics, 14(19), 3929. DOI: 10.3390/electronics14193929. 8.      Zhang, R., Luo, X., He, J., Niyato, D., Kang, J., Xiong, Z., & Li, Y. (2026). Compact LLM Deployment and World Model Assisted Offloading in Mobile Edge Computing. arXiv:2602.13628. 9.      Yang, W., & Xia, S. (2026). PELM: Power Efficient On-Device LLM Inference with Speculative Decoding and Dynamic Voltage Frequency Scaling. SenSys 2026, 438–451. DOI: 10.1145/3774906.3802783.