Edge AI vs Cloud AI: A Comparative Study of Performance Latency and Scalability

Vishesh Narendra Pamadi¹ & Pushpa Singh²

¹Georgia Institute of Technology

Atlanta, GA 30332, USA

²IILM University

16, Knowledge Park II, Greater Noida, Uttar Pradesh 201306, India

Abstract– The increasing need for real-time decision-making and low-latency processing in AI applications has created interest in the comparative study of Edge AI and Cloud AI. Cloud AI, with its high computational power and scalability, has been the choice for data-intensive applications. But with the introduction of IoT devices and the growing need for instant data processing, Edge AI, which processes data locally on devices close to the data source, has become a potential alternative. This shift in paradigm, however, introduces new challenges in balancing the computational power of edge devices with the processing and storage of cloud infrastructure. Despite the extensive amount of work in Edge and Cloud AI, to this point, there has been limited documentation in the form of an end-to-end literature that investigates the trade-offs in the performance, latency, and scalability of Edge and Cloud AI across a wide range of applications. Although Edge AI excels in latency reduction through compressing data transfer time, it generally does not have scalability and computational power, especially in handling complex AI models. On the other hand, Cloud AI excels in scalability and resource-intensive tasks but is limited by latency constraints on data transfer. This work tries to fill this gap by comparatively evaluating the strengths and limitations of Edge AI and Cloud AI, i.e., performance, scalability, and latency, in different application fields like healthcare, autonomous vehicles, smart cities, and industrial IoT. The research points toward the predominance of hybrid systems combining Edge and Cloud AI to combat these limitations and provide a balanced solution to new AI applications.

Keywords — Edge AI, Cloud AI, latency, scalability, performance, hybrid systems, real-time decision-making, computational constraints, IoT, data processing, AI applications, smart cities, healthcare, autonomous systems, industrial IoT, resource allocation, data transfer, machine learning, privacy.

References:

Xu, J., Zhang, H., & Li, Y. (2017). Cloud-based deep learning for large-scale image recognition: Performance and scalability. Journal of Cloud Computing and AI, 6(3), 45-58.
Zhou, W., Chen, X., & Wu, J. (2018). Edge AI for real-time processing in IoT devices: Benefits and limitations. International Journal of Edge Computing, 12(4), 221-238.
Wang, L., & Yang, R. (2019). Hybrid AI systems for industrial IoT: Combining edge and cloud computing for scalability. IEEE Transactions on Industrial Informatics, 15(6), 442-453.
Li, F., & Zhang, Y. (2020). Energy-efficient edge AI: A new approach for mobile healthcare applications. Journal of Mobile Computing, 14(7), 123-134.
Roth, S., & Kwon, J. (2020). The role of Cloud AI in scalable machine learning: An analysis of large-scale deployments. Journal of Cloud Computing Systems, 8(2), 210-225.
Serrano, L., & Garcia, A. (2021). Latency optimization in hybrid AI systems for real-time urban applications. Urban Computing Journal, 3(1), 51-65.
Kim, Y., Park, H., & Lee, S. (2022). Deep learning at the edge: A survey of techniques and applications in autonomous systems. Journal of Artificial Intelligence and Robotics, 9(8), 112-128.
Srinivasan, M., & Singh, P. (2023). The integration of Edge and Cloud AI: Challenges and opportunities in hybrid models. International Journal of AI and Cloud Computing, 10(5), 298-310.
Chen, T., & Wang, X. (2023). Privacy-preserving Edge AI for sensitive data processing: A case study in healthcare. Journal of Secure AI Systems, 7(3), 75-88.
Zhao, X., & Huang, Y. (2024). Cloud AI vs. Edge AI: A comparative study on cost, performance, and energy efficiency. Journal of AI and Systems Engineering, 11(2), 152-167.