The Efficiency of Distributed Cloud Computing in AI Models

Journal: Journal of Artificial Intelligence, Machine Learning and Data Science
Authors: Srinivasa Kalyan Vangibhurathachhi
DOI: https://doi.org/10.51219/JAIMLD/srinivasa-kalyan-vangibhurathachhi/542

Abstract

The rapid advancements in Artificial Intelligence (AI) have led to increased computational demands, particularly for deep neural networks (DNNs) and large language models (LLMs). Traditional centralized cloud computing struggles to efficiently support AI workloads due to high latency, bandwidth constraints, energy inefficiencies and scalability limitations. To address these challenges, Distributed Cloud Computing (DCC) has emerged as an innovative platform that decentralizes computational resources, bringing them closer to data sources to enhance efficiency, reduce costs and optimize performance. This paper critically examines the efficiency of DCC in AI models.

Through decentralized resource allocation, optimized workload distribution and real-time data processing, DCC offers benefits such as lower latency, improved computational scalability, energy efficiency, resource utilisation, data privacy and compliance. A comparative analysis with centralized and edge cloud models highlights the advantages of DCC, particularly in latency reduction, high scalability, throughput and energy efficiency. Additionally, key use cases in autonomous vehicles, healthcare, finance and retail highlight DCC’s transformative potential across industries. Further, future trends in DCC for AI models are focussed on; quantum cloud computing for AI model acceleration, integration of 5G in distributed cloud systems and AI-drive distribute cloud orchestration. Despite its benefits, security vulnerabilities, workload management complexities and significant computational resources which are needed remain key challenges. Addressing these issues will be crucial for maximizing AI efficiency in the era of distributed computing.

Keywords: Distributed cloud computing (DCC), edge computing, federated learning, centralised cloud infrastructure, AI model, efficiency.

1. Introduction
Artificial intelligence is transforming industries from finance and healthcare to entertainment, autonomous systems and smart cities. However, the current AI models such as deep neural networks (DNNs) and large language models (LLMs) demand massive computational resources for training and real-time inference¹. To address the growing computational demands of AI applications, the integration of AI with cloud computing has emerged as a promising approach². By leveraging the scalability, cost-effectiveness, flexibility and collaborative opportunities provided by cloud computing, practitioners and researchers can develop and deploy innovative AI solutions across a wide range of domains for various industries.

However, Abughazalah, et al.³ noted that the traditional centralized cloud infrastructures present challenges such as latency, cost, bandwidth limitations and energy inefficiencies especially when handling massive datasets required for training complex AI models. In addressing these issues, distributed cloud computing (DCC) has emerged as a promising paradigm that decentralises data processing by bringing computational resources closer to data sources⁴. Unlike traditional cloud models that centralizes resources, DCC focuses on dispersing resources to specific data centre locations to enhance performance and scalability while reducing latency, bandwidth constraints and operational costs⁴. Leading companies such as Amazon Web Services, Google Cloud and Microsoft Azure have developed and adopted distributed architectures to enhance AI model efficiency and scalability².

To this end, this research paper explores the efficiency of distributed cloud computing in AI models by;

Evaluating the efficiency of DCC in AI workloads
Comparing DCC performance with centralized and hybrid computing architectures.
Analysing its impact on cost, latency, energy consumption and scalability
Identifying key challenges and proposing future directions for distributed cloud computing in AI models.

2. Problem Statement
In the views of Zangana, et al⁵, the traditional AI computing models rely on centralized cloud data centres which face significant bottlenecks when processing large scale AI workloads. For instance, the need to transfer vast amounts of data to centralized data centres results in high latency and bandwidth consumption, thus affecting the performance of AI applications particularly for those requiring real-time processing like internet of things and autonomous vehicles. Abughazalah, et al.³ adds that centralized AI systems require large-scale data centres which involve high capital and operational expenses. In case of technical problems, centralized systems can become single points of failure thus raising concerns about reliability and resilience. Moreover, the energy demands of large data centres substantially to the operational costs and carbon footprints of centralised data centres⁶.

While traditional AI computing present significant bottlenecks, challenges also exist in AI model training and deployment. Nama⁷ argues that training AI models requires considerable computational power and storage capacities. Accordingly, centralized cloud systems may struggle to efficiently allocate resources for parallel processing tasks, leading to suboptimal performance. Additionally, the data privacy and security are critical concerns when transferring sensitive information to centralized locations due to the increased risk of breaches. Medium⁸ reports that ensuring compliance with data protection regulations in various countries and regions further complicates the training and deployment of centralized AI models. To overcome these limitations, DCC which enables AI workloads to be executed across multiple cloud environments, hybrid clouds and edge nodes is crucial⁹. This improves performance, cost-effectiveness and sustainability while reducing dependency on single-location processing.

Figure 1: Challenges in AI model training (Tikong, 2024).

Figure 2: Challenges in AI model deployment (Sinha & Lee, 2024).

3. Proposed Solution: Distributed Cloud Computing for AI Models Efficiency

3.1. What is DCC

By definition, DCC refers to the decentralized allocation of computational resources across multiple data centres including public, private, hybrid and edge clouds¹⁰. With this targeted and centrally managed distribution of public cloud services, businesses can deploy and run applications in a mix of cloud environments that meet their requirements for performance, regulatory compliance and much more. Ramamoorthi¹¹ highlights that DCC resolves the operational and management inconsistencies which occur in centralised cloud or multicolour environments. Most importantly, DCC offers the ideal foundation for edge computing which involves running servers and applications closer to where data is created¹². Key technologies have enabled DCC include; containerization and orchestration¹³ and serverless AI computing technologies^14,15. Other technologies that are useful in DCC are; Federated AI and decentralized learning (TensorFlow Federated, PySyft)¹⁶ and AI-optimized hardware accelerators (Tensor Processing Units, Field Programmable Gate Arrays, Graphics Processing Units clusters)¹⁷.

Figure 3: DCC (Abdi & Zeebaree, 2024).

3.2. How distributed cloud computing enhances AI efficiency

Extant literatures and empirical studies indicate that DCC can improve the efficiency of AI models through various ways. Ramamoorthi¹¹ highlight that DCC reduces latency in AI applications such as those requiring real-time processing like autonomous vehicles and financial fraud detection. Although traditional cloud-based AI processing requires sending data to centralized servers which leads to latency issues, distributing workloads closer to data sources through edge computing significantly reduces processing times¹¹. Sengupta¹⁸ reports that Google's Edge TPU technology utilizes distributed cloud frameworks to process AI workloads closer to the source thus reducing delays in AI applications like speech recognition and real-time video analytics.

DCC systems can improve the efficiency of AI models through optimised resource allocation. Although traditional cloud models tend to suffer from inefficient resource allocation where certain data centres remain underutilized, distributed systems addresses this issue by allocating computational tasks across various nodes within the network based on demand¹⁹. Such an approach maximizes performance of individual nodes and enhances the responsiveness and efficiency of systems. This evident in AWS where Lambda and Fargate offer serverless computing by allowing AI workloads to be distributed dynamically without requiring dedicated resources thus reducing operational costs¹⁴.

Abdi and Zeebaree¹⁹ adds that DCC can improve AI efficiency through parallel processing where computational workloads are distributed across multiple nodes. This enhances the efficiency of model training by leveraging parallelism in computation-heavy tasks such as deep learning-based image classification or NLP model training. By leveraging capabilities of several computers, parallel processing effectively addresses the computational demands of intricate and resource-intensive tasks. Youvan cited OpenAI’s GPT-4 model training which uses distributed computing across multiple data centres to optimize parallel processing thus reducing training time compared to single-node computations.

Differently, Nama⁷ points out that DCC improves AI efficiency through reduction in energy consumption. While centralised data centres consumers’ massive amounts of energy, distributed AI models can leverage edge computing to significantly reduce power consumption. A study by Huang, et al²⁰. indicated that DCC can reduce power consumption by between 19%-28% through reducing the need for constant data transfers. Another way in which distributed systems enhances efficiency of AI models is through scalability which enables efficient management of increased workload. By augmenting the distributed systems with additional nodes as needed, Khan²¹ noted that distributed systems remain capable of handling increasing computational demands. Vertical scaling (increasing resources of nodes) and horizontal scaling (adding more nodes) are common approaches to accommodating increased workload and maintaining operational efficiency²². Scalability is exhibited by Microsoft Azure’s AutoML platform which uses distributed computing to dynamically scale AI models training across several cloud regions thus enhancing the processing power for complex machine learning tasks²³. However, poorly controlled scaling may reduce overall performance of AI models²².

More important, distributed systems can enhance AI model efficiency and performance through data privacy and compliance by processing extensive data within a specific region rather than transferring them to centralised cloud servers². Given that data privacy regulations such as GDPR and CCPA impose strict controls on data processing and storage, distributed cloud servers help in this. Through federated learning, a distributed AI approach, Thakur²⁴ noted that AI models are trained across multiple and decentralized devices while keeping data local, enhancing privacy and reducing computational burden on servers. While distributed systems can enhance data privacy, Chavan and Raut²⁵ argued that it introduces vulnerabilities such as data fragmentation and multi-point attack risks. Ensuring secure communication between distributed nodes remains a key challenge for tech companies.

4. Used Cases of DCC In AI Applications
Several case studies that demonstrate how DCC is used in AI models to unlock new capabilities and enhance efficiency for businesses and industries. A notable example is Netflix which uses distributed cloud-based AI to provide personalised content recommendations for millions of users around the world ²¹. By leveraging scalability of cloud computing, Netflix analyses large amounts of user data in real-time to fine tune recommendation algorithms thus delivering highly personalized experience to users. Another successful case of DCC in AI applications is in the development of autonomous vehicle where automakers like Tesla utilise cloud-based AI to analyse data from self-driving cars, train AI models and improve the performance of autonomous systems²¹. These cases demonstrate that DCC offers suitable infrastructure for processing large datasets as AI drives automation and intelligent decision-making.

Additionally, integration of DCC in AI models is reshaping industries by developing innovative services, improving decision making and enhancing new efficiencies²¹. In healthcare sector, AI-based cloud platforms have revolutionised patient care and diagnostics. For example, IBM Watson Health deploys AI in distributed IBM Bluemix Clouds for analysing medical records and suggesting treatment options whereas cloud-based AI applications are utilised to predict disease outbreaks and improve hospital resource management²⁶.

In the finance industry, distributed clouds and AI applications are enhancing customer service, fraud detection and risk management. JP Morgan exemplifies this by utilising AI-powered cloud platforms to determine fraudulent transactions in real-time and robot advisors to suggest personalised financial advice²⁷. In the retail sector, DCC and AI models are revolutionising innovations like personalised marketing, dynamic pricing and optimisation of inventories. Amazon has deployed cloud-based AI applications to process consumer preferences, forecast supply chain demands and suggest personalized product recommendations effectively²⁸.

5. Impact Analysis: Efficiency Metrics and Comparative Evaluations

5.1. Performance metrics for evaluating DCC in AI models

The effectiveness of DCC in AI models can be evaluated based on;

Latency reduction: Distributed systems significantly reduce latency by distributing computational tasks closer to data sources. Can be measured through response time (time taken for an AI model to process input and return an output) and inference time (how quickly AI model can process and predict outcomes in real-time application) (Naveen et al., 2021).

Figure 4: reduced latency due to edge computing²⁹.

Computational throughput: Measures how many AI tasks can be processed simultaneously, directly impacting scalability. AI training models require massive parallel computations which are measured in petaFLOPS (Floating Point Operations per Second. Distributed systems allow AI models to run simultaneously across several nodes, thus increasing throughput³⁰.
Energy consumption: DCC lowers power requirements compared to centralized cloud computing and can be measured in Power Usage Effectiveness (Measures how efficiently data centre uses energy)³¹.
Cost-effectiveness: DCC reduces operational expenses through serverless and dynamic resource allocation. Cost effectiveness can be measured in pay-per-use model where users only pay for the actual compute time consumed by code. This means no more paying for idle servers or underutilized resources³².

5.2. Comparative analysis
A summarised evaluation of DCC AI versus centralised cloud computing AI and hybrid computing cloud AI is presented as follows;

Table 1: Adapted^19,21,11,2.

Metric	Centralized Cloud AI	Distributed Cloud AI	Hybrid computing AI
Latency	High	Low	Lowest
Scalability	Limited	High	Moderate
Training Cost	High	Moderate	High (hardware costs)
Data Privacy and security	Low	High (federated learning)	High
Throughput (FLOPS)	1-10 PFLOPS	10-100 PFLOPS	5-50 PFLOPS
Energy Efficiency	Low	High	Moderate

6. Scope and Challenges of DCC For AI

6.1. Future trends and research directions

The landscape of DCC is continuously evolving with key trends shaping the technological advancement. A key future trend is the convergence of quantum computing and distributed cloud which will significantly accelerate AI model training and inference especially for tasks requiring high-dimensional optimization and massive parallelism³³. This will help in faster training of deep learning models such as GPT-5 and beyond as well as optimize neural network weights using quantum-enhanced optimizers. Currently IBM is testing Quantum Cloud AI for distributed deep learning acceleration to determine exponential speedups in AI model processing³⁴. Another key trend is the increasing use of AI to optimise distributed cloud orchestration which enables dynamic resource allocation, workload balancing and auto-scaling of AI models within distributed cloud environments. An example is Microsoft’s Azure AI which uses reinforcement learning to optimize virtual machine allocation thus improving cloud performance²³. Further, Nassef, et al.³⁵ reported that the emergence of 5G networks will enhance distributed cloud-based AI model performance by enabling low latency data processing at the edge thereby reducing dependency on central cloud servers. This will impact real-time AI inference for IoT applications in smart cities, retail, autonomous vehicles and much more. It will also lower data transmission costs by processing AI workloads closer to the source. By understanding these trends in distributed computing organizations can better harness its potential for scalable and efficient AI model training.

The future research directions should focus on federated learning which has gained significant attention as a method to train AI models across multiple edge devices without sharing raw data, thus addressing privacy concerns²⁴. The research should seek to address challenges in communication overhead within large-scale federated learning networks and ensure model consistency across decentralized training nodes. Verdecchia, et al.³⁶ adds that future researchers should explore the environmental impact of AI model training which has led to the rise of Green AI by focusing on reducing carbon footprints in distributed cloud computing. Key areas of focus should be on energy-efficient AI algorithms that require fewer computations and renewable energy-powered cloud data centres to offset emissions. As AI models are deployed across multi-cloud environments, Jalal et al. observed that future researchers should examine the interoperability issues which arise by requiring standardization in model deployment, deep learning, APIs and data formats. Specifically, researchers should consider Open Neural Network Exchange (ONNX) which has emerged as a standard framework to ensure AI model portability across different cloud providers.

6.2. Challenges
Despite the advantages, AI models in distributed cloud face vulnerabilities such as data leakage, adversarial attacks and model inversion attacks⁵. The security may arise from data poisoning attacks where malicious nodes can inject false data into federated learning models or adversarial AI hacking where attackers manipulate AI model predictions through small perturbations. Baranwal, et al³⁷. indicated that privacy and security issues can be resolved through blockchain-driven AI security which ensures tamper proof federated learning. Zangana, et al⁵. points out complexities in managing distributed AI workload across several cloud nodes which introduces data synchronization issues, scheduling inefficiencies and increased orchestration complexity. This challenge can be resolved by AI-based predictive workload scheduling to optimize cloud resource allocation. Further, Yuan and Zhou³⁸ pointed that deploying high-performance AI models within distributed cloud environments requires significant computational resources, which may not be accessible to all enterprises. Decentralizing AI cloud platforms using blockchain and shared computational power can help in resolving this issue. Table 2 below summarizes the key challenges and research opportunities in DCC.

Table 2: DCC challenges¹¹.

7. Conclusion
Through decentralized resource allocation, optimized workload distribution and real-time data processing, DCC offers benefits such as lower latency, improved computational scalability, energy efficiency, resource utilization, data privacy and compliance. The study reveals that tech companies like AWS, Google Cloud and Microsoft Azure have increasingly adopted distributed architectures to improve AI model performance and cost-effectiveness. Additionally, key use cases in autonomous vehicles, healthcare, finance and retail highlight DCC’s transformative potential across industries.

However, several challenges persist, including security vulnerabilities, interoperability issues and workload management complexities. AI models deployed in distributed environments face threats like data leakage, adversarial attacks and compliance risks and significant computational resources which requires solutions. Future research should focus on federated learning advancements, green AI initiatives and quantum-enhanced AI training to further improve DCC and AI model efficiency. As AI continues to evolve, the convergence of DCC with emerging technologies such as 5G, edge computing and quantum computing will shape the future of AI model training and deployment. Overcoming the current challenges will be key in maximizing DCC's potential as the DNA of AI model’s infrastructure thus enabling faster, more secure and scalable AI solutions across various domains.

8. References

1. Hagos DH, Battle R and Rawat DB. Recent advances in generative AI and large language models: Current status, challenges and perspectives. IEEE Transactions on Artificial Intelligence.

2. Mungoli N. Scalable, distributed AI frameworks: leveraging cloud computing for enhanced deep learning performance and efficiency.

3. Abughazalah M, Alsaggaf W, Saifuddin S and Sarhan S. Centralized vs. Decentralized Cloud Computing in Healthcare. Applied Sciences, 2024;14: 7765.

4. Samunnisa K, Kumar GSV and Madhavi K. Intrusion detection system in DCC: Hybrid clustering and classification methods. Measurement: Sensors, 2023;25: 1-12.

5. Zangana HM, khalid Mohammed AK and Zeebaree SR. Systematic review of decentralized and collaborative computing models in cloud architectures for distributed edge computing. Sistemasi: Jurnal Sistem Informasi, 2024;3: 1501-1509.

6. https://www.restack.io/p/centralized-vs-decentralized-ai-answer-2023-articles

7. Nama P. Integrating AI with cloud computing: A framework for scalable and intelligent data processing in distributed environments. International Journal of Science and Research Archive, 2022;6: 280-291.

8. https://medium.com/coinmonks/decentralized-ai-vs-centralized-ai-key-differences-and-advantages-3bff25589782

9. Rashid ZN, Zebari SR, Sharif KH and Jacksi K. DCC and distributed parallel computing: A review. In 2018 International Conference on Advanced Science and Engineering (ICOASE), 2018: 167-172.

10. https://www.ibm.com/think/topics/distributed-cloud

11. Ramamoorthi V. Exploring AI-Driven Cloud-Edge Orchestration for IoT Applications. International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 2023;5: 385-393.

12. Vankayalapati RK, Sondinti LRK, Kalisetty S and Valiki S. Unifying Edge and Cloud Computing: A Framework for Distributed AI and Real-Time Processing. Journal for Re Attach Therapy and Developmental Diversities, 2023;69: 1913-1926.

13. Hardikar S, Ahirwar P and Rajan S. Containerization: Cloud Computing based Inspiration Technology for Adoption through Docker and Kubernetes. 2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC), Coimbatore, India, 2021: 1996-2003.

14. https://aws.amazon.com/serverless/

15. Malawski M, Gajek A, Zima A, et al. Serverless execution of scientific workflows: Experiments with hyperflow, AWS lambda and google cloud functions. Future Generation Computer Systems, 2020;110: 502-514.

16. Chelliah PR, Rahmani AM, Colby R, et al. Model Optimization Methods for Efficient and Edge AI: Federated Learning Architectures, Frameworks and Applications. John Wiley & Sons, 2024.

17. Hambali Y, Riwurohi JE and Akbar V. Smart Strategies in Hardware Provisioning for Ai Solutions in The Cloud. Eduvest-Journal of Universal Studies, 2024;4: 11387-11399.

18. Sengupta J, Kubendran R, Neftci E, et al. High-speed, real-time, spike-based object tracking and path prediction on google edge TPU. International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2020: 134-135.

19. Abdi AS, Zeebaree SRM. Embracing Distributed Systems for Efficient Cloud Resource Management: A Review of Techniques and Methodologies. Indonesian Journal of Computer Science, 2024;13: 1912-1933.

20. Huang T, Huang W, Zhang B, et al. Optimizing energy consumption in centralized and distributed cloud architectures with a comparative study to increase stability and efficiency. Energy and Buildings, 2025;333: 115454.

21. Khan F. Artificial Intelligence and Cloud Computing: A Perfect Symbiosis. Advances in Computer Sciences, 2023;6: 1-12.

22. Malhotra S. The Current State of Research into the Efficiency of Distributed Machine Learning Algorithms for Cloud-Based Big Data Analysis. International Journal of Advance Research, Ideas and Innovations in Technology, 2025;11: 132-134.

23. https://azure.microsoft.com/en-us/products/machine-learning/

24. Thakur D. Federated Learning and Privacy-Preserving AI: Challenges and Solutions in Distributed Machine Learning. International Journal of All Research Education and Scientific Methods (IJARESM), 2021;9: 3763-3764.

25. Chavan MR and Raut SY. Cloud Security Solution: Fragmentation and Replication. Communication of the ACM, 2013;56: 64-73.

26. Aggarwal M and Madhukar M. IBM’s Watson analytics for health care. Cloud Computing Systems and Applications in Healthcare, 2017: 117-134.

27. Praveen RVS. Banking in the Cloud: Leveraging AI for Financial Transformation. Addition Publishing House, 2024.

28. Anbalagan K. AI in Cloud Computing: Enhancing Services And Performance. International Journal of Computer Engineering And Technology (IJCET), 2024;15: 622-635.

29. Kolhar M, Al-Turjman F, Alameen A and Abualhaj MM. A three layered decentralized IoT biometric architecture for city lockdown during COVID-19 outbreak. Ieee Access, 2020;8: 163608-163617.

30. Shankar S and Reuther A. Trends in energy estimates for computing in ai/machine learning accelerators, supercomputers and compute-intensive applications. In 2022 IEEE High Performance Extreme Computing Conference, 2022: 1-8.

31. https://www.grcooling.com/blog/power-usage-effectiveness-a-simple-guide-to-improving-your-data-centers-energy-consumption/#:~:text=The%20power%20usage%20effectiveness%20measures%20a%20data,affects%20profitability%2C%20computational%20performance%2C%20and%20the%20environment

32. https://techdynasty.medium.com/diving-deep-into-cost-effectiveness-of-serverless-computing-on-aws-b0acbe86e9df

33. Hossain MI, Sumon SA, Hasan HM. Quantum-Edge Cloud Computing: A Future Paradigm for IoT Applications, 2024.

34. https://www.forbes.com/sites/tiriasresearch/2024/06/24/ibm-develops-the-ai-quantum-link/