CLOUD REPATRIATION: AN EMPIRICAL ANALYSIS OF WHEN MOVING BACK FROM CLOUD CREATES BUSINESS VALUE

ABSTRACT

The article is devoted to the study of cloud workload repatriation as a direction for optimizing digital infrastructure under changing economic and architectural parameters of corporate system operation. The paper reveals the prerequisites for the transition from universal cloud strategies to more differentiated models of digital workload placement, as well as the factors influencing the creation of positive business value when systems are moved back from the public cloud. A methodological approach to assessing the feasibility of repatriation is proposed, based on the use of architectural analysis data, economic evaluation of alternatives, the cloud value decay model, and the cloud repatriation suitability index. It is specified that repatriation decisions should be made with regard to workload predictability, the degree of dependence on cloud-native services, the scale of costs, performance requirements, and constraints related to data and regulation. The findings conclude that repatriation is most appropriate for mature, stable, and resource-intensive workloads, whereas hybrid and cloud deployment models remain more effective for systems with mixed or highly variable workload profiles.

АННОТАЦИЯ

Статья посвящена исследованию репатриации облачных нагрузок как направления оптимизации цифровой инфраструктуры в условиях изменения экономических и архитектурных параметров эксплуатации корпоративных систем. В работе раскрыты предпосылки перехода от универсальных облачных стратегий к более дифференцированным моделям размещения цифровых нагрузок, а также факторы, которые влияют на формирование положительной бизнес-ценности при возврате систем из публичного облака. Предложен методический подход к оценке целесообразности репатриации, основанный на применении данных архитектурного анализа, экономической оценки альтернатив, модели убывания ценности облака и индекса пригодности нагрузки к репатриации. Уточнено, что принятие решения о репатриации должно осуществляться с учетом предсказуемости нагрузки, степени зависимости от cloud-native-сервисов, масштаба затрат, требований к производительности и ограничений, которые связаны с данными и регулированием. По итогам исследования установлено, что наибольшая целесообразность репатриации характерна для зрелых, стабильных и ресурсоемких нагрузок, в то время как для систем со смешанным или высоковариативным профилем более эффективными остаются гибридные и облачные модели размещения.

Keywords: cloud workload repatriation, cloud computing, digital infrastructure, business value, architectural analysis, hybrid model, cloud-native services, cost optimization, public cloud, corporate systems.

Ключевые слова: репатриация облачных нагрузок, облачные вычисления, цифровая инфраструктура, бизнес-ценность, архитектурный анализ, гибридная модель, cloud-native-сервисы, оптимизация затрат, публичное облако, корпоративные системы.

Introduction. The contemporary context shaped by the development of the digital economy, platform-based business models, and distributed IT environments has led to the widespread adoption of cloud computing as a foundational infrastructure model for hosting corporate systems. At the early stages of digital transformation, the cloud environment provided organizations with rapid access to computing resources, enabling their flexible scaling, accelerating time-to-market, and reducing initial entry barriers. As noted in the scholarly literature addressing the rationale for the adoption of cloud computing, the cloud can serve as a technological foundation for the development of information systems [1]. From a regulatory perspective, cloud computing is defined as a model for providing on-demand access to a shared pool of configurable resources, which establishes it as a distinct form of organizing computing infrastructure [10].

However, as practical experience with cloud solutions has accumulated, the focus of both researchers and businesses has shifted from the general recognition of cloud advantages to a more precise assessment of their economic and architectural efficiency in relation to specific workload types. For organizations operating predominantly mature, resource-intensive, and predictable digital systems, particular importance has been placed on issues such as total cost of ownership, vendor dependence, transparency of operational expenditures, performance, and compliance with industry requirements. In this context, cloud workload repatriation—i.e., the migration of systems previously hosted in the public cloud back to private, on-premises, or colocation infrastructure—is increasingly emerging as a promising direction.

The relevance of the topic under consideration is determined by the fact that cloud workload repatriation reflects a gradual transition toward more specific and finely tuned strategies for digital system placement. According to industry reports, a significant proportion of companies already employ hybrid models (combining on-premises and cloud resources) or are evaluating the feasibility of repatriating certain workloads from the public cloud in the presence of pronounced economic or operational constraints [4]. Practical interest in this issue is further reinforced by observable industry practices, as reconsideration of the cloud model is often accompanied by cost reductions and the realization of additional benefits [12]. Accordingly, the scientific significance of this study lies in the need to clarify the conditions under which repatriation genuinely generates positive business value and becomes justified. While the contemporary literature provides a detailed account of the advantages of cloud computing, considerably less attention has been paid to the point at which the cloud model for a specific workload loses its initial efficiency and viability. In this regard, there is a growing need to develop analytical tools that enable the evaluation of infrastructure decisions through the lens of workload characteristics, the degree of dependence on cloud services, cost scale, performance requirements, and organizational and regulatory constraints. All of the above has defined the objective of this study.

The aim of this study is to substantiate a methodological approach to assessing the feasibility of cloud workload repatriation and to identify the factors under which migration from the public cloud contributes to improving the economic and architectural efficiency of a digital system.

Methodology of the study. The methodological framework of the study is based on architectural analysis, the economic evaluation of alternatives, and the comparative analysis of digital workload placement scenarios. Within this framework, cloud workload repatriation is interpreted as a managerial and architectural decision directly associated with the dynamics of IT infrastructure value throughout the system lifecycle.

From the perspective of theoretical foundations, the first approach is based on the consideration of the total cost of ownership of both cloud and non-cloud solutions; according to this approach, the choice of an infrastructure model should account for long-term operational costs associated with computing, storage, network traffic, maintenance, administration, and data transfer [7].

The analysis of architectural alternatives implies that an infrastructure decision is evaluated through a system of trade-offs between quality attributes (performance, reliability, manageability, scalability, and security) and cost. When choosing between retaining workloads in the public cloud, partially repatriating them, or migrating them to on-premises infrastructure, it is advisable to consider the economic implications of each option [2]. In this regard, approaches to architectural evaluation are also of particular importance, as they enable the structuring of trade-offs and the alignment of system technical characteristics with organizational business constraints [8].

For the purpose of assessing the level of cloud-native dependency, it is taken into account that modern systems are often designed as cloud-native applications based on managed services, orchestration, containerization, event-driven architecture, and serverless approaches. This ensures development speed and preserves flexibility in product evolution scenarios, while simultaneously affecting the cost and complexity of reverse migration. Studies of cloud-native environments emphasize that architectural dependence on specific cloud services is one of the key factors influencing the boundaries of an organization’s technological freedom [6].

At the same time, in practice it is advisable to consider intermediate deployment options, as for most organizations the choice is not between two extremes but among several configurations, including the public cloud, hybrid models, multi-cloud deployment, and private infrastructure [5].

The empirical basis of the study comprises publications on repatriation practices, industry reports, case studies of companies with disclosed outcomes of cloud strategy reassessment, as well as materials enabling the comparison of workload types, their architectural implementation, and the resulting economic effects. To assess the applicability of repatriation, the following analytical parameters are used: workload predictability, the scale of sustained resource consumption, the degree of dependence on proprietary cloud services, latency requirements and data control, as well as the presence of a significant cost differential between the cloud and alternative deployment models.

In addition, a methodological approach is proposed within the framework of this study. The assessment of the feasibility of cloud workload repatriation is complemented by formalized instruments: (1) the Cloud Value Decay Model (CVDM) and (2) the Cloud Repatriation Suitability Index (CRSI). The first model makes it possible to determine the point at which cloud deployment transitions from a zone of positive to negative business value, while the second enables a quantitative assessment of the likelihood of successful repatriation for a given workload. Each of these instruments is considered separately below.

Cloud Value Decay Model (CVDM). The basic equation of the model is defined as follows:

where:  – time of workload operation in the cloud;   – the aggregate benefits of cloud deployment;   – the total costs of using cloud infrastructure;   – the accumulated “complexity tax” associated with managing architecture, dependencies, and operations.

The model is interpreted as follows:

‒  when   the cloud continues to deliver positive business value;

‒ when  an architectural and economic inflection point is reached;

‒ when  cloud deployment begins to reduce the overall efficiency of the system, and repatriation becomes a justified option.

Within the framework of the model, the aggregate benefits are defined as the sum of three components:

At the same time, the effect of adaptability and speed of deployment diminishes as the system matures and can be represented as an exponential function:

where  – is the initial value of infrastructure flexibility,  – is the coefficient of its decline over time. In the model under consideration, for mature predictable workloads, the parameter is set at   per year.

The scalability effect is determined through workload variability:

where:

The closer the peak load is to the average load, the lower the additional value of cloud elasticity. For predictable workloads, the value of this component approaches zero.

Additionally, a function of accumulated architectural complexity is introduced:

where   represents the baseline operational costs of cloud maintenance,  – the annual increase in complexity.  In this study, this effect is interpreted in terms of the growth of overhead costs associated with maintenance, observability, dependency management, and overcoming vendor lock-in.

The second instrument—the Cloud Repatriation Suitability Index (CRSI)—is constructed as a composite index:

In the index, all key parameters are normalized within the range from 0 to 100, while an additional adjustment factor reflects the relative “cloud premium.”

The adjustment coefficient is calculated as follows:

or, in expanded form,

Thus, if the current cloud TCO is twice as high as the estimated non-cloud TCO, the index receives an additional 25 points; in the case of a threefold difference, 50 points are assigned.

Workload predictability is assessed as follows:

where

The greater the workload variability, the lower the resulting predictability score.

A low level of dependence on cloud-native services is represented as follows:

where  – the number of critical services without equivalent alternatives in non-cloud environments.

The scale of cloud expenditures is assessed using the following formula:

Accordingly, the higher the monthly cloud expenditures, the greater the economic rationale for undertaking repatriation.

Performance requirements are assessed as follows:

The indicator reflects the sensitivity of the workload to latency and throughput.

Regulatory constraints are taken into account as follows:

The index increases in the presence of stringent requirements for data localization, industry regulation, and infrastructure control.

The interpretation of the resulting index value implies the distribution of values across defined ranges:

‒  – high suitability for repatriation;

‒  – moderate suitability; a hybrid scenario or additional assessment is recommended;

‒  – low suitability for repatriation; cloud deployment retains greater value.

For a mature and predictable workload with a stable scale of operation, the model indicates that as the effect of infrastructure adaptability (its flexibility) diminishes and operational complexity accumulates, the overall cloud value may transition into the negative range within 18–36 months after achieving a pronounced scale effect. The following illustrative calculation example can be provided:

Then, at   years:

The resulting negative value indicates that, for a mature and predictable workload, the cloud model in this case ceases to generate positive business value and, therefore, may be considered within the context of repatriation.

Results and Discussion. The conducted study demonstrates that cloud workload repatriation represents a natural form of digital infrastructure optimization in cases where the actual operational parameters of a system begin to differ significantly from the conditions under which the cloud model provided maximum efficiency.

Thus, at the early stages of the digital product lifecycle, deployment in the public cloud accelerates service rollout, simplifies scaling, and reduces the need for significant upfront capital investment. However, as workloads stabilize, the volume of sustained resource consumption increases, and requirements for architectural control become more complex, the relative importance of these advantages gradually shifts, with total cost of ownership, performance, manageability, and the predictability of operational expenses becoming decisive factors.

The analysis demonstrates that for mature digital systems, the cloud environment does not always maintain the same level of economic efficiency throughout the entire lifecycle. At the launch and growth stages, the cloud enables rapid responses to changing demand; however, under conditions of stable workload profiles and a high share of continuously utilized resources, its economic justification becomes less evident. This is due to the fact that ongoing expenses for computing, data storage, network traffic, and managed services increase the overall fixed burden on the budget, while the additional benefits of elasticity are only partially realized. According to reviews of current cloud strategy practices, a significant proportion of organizations are already transitioning toward more differentiated deployment models, in which the public cloud is used selectively, typically in combination with private or colocation infrastructure [4; 9].

Therefore, the value of cloud deployment should be considered a variable that depends on the type of workload, the scale of the system, and its stage of maturity. In practice, this implies that the same infrastructure model may be effective for certain workloads and less appropriate for others. This condition is particularly evident in systems with predictable resource consumption profiles, high throughput requirements, and relatively low dependence on specialized cloud services, as for such solutions the repatriation of part of the workload to on-premises infrastructure enables greater control over the computing environment and a more precise alignment of architectural configuration with long-term operational parameters.

This relationship can be visually represented as follows (Figure 1).

Figure 1. Dynamics of the decline in the business value of cloud deployment for a mature, predictable workload, compiled by the author

As can be observed, at the initial stage of operation the cloud model is characterized by a high level of value due to rapid scalability; however, as the system transitions to a stage of stable operation, the curve of overall utility gradually declines. At a certain point, a threshold is reached beyond which continued use of the public cloud no longer ensures an optimal alignment between costs and the resulting benefits. It is within this zone that the likelihood increases that partial or full repatriation will represent a rational direction for further optimization.

The study demonstrates that the primary practical interest lies not in merely identifying the fact of migration from the cloud, but in determining the types of workloads for which such a transformation genuinely creates a positive effect. In this regard, it is advisable to distinguish three main groups (Table 1).

1. This group includes workloads that retain a high degree of dependence on the public cloud, encompassing systems with highly variable resource consumption, pronounced seasonality, a need for rapid global scaling, active experimentation, and tight integration with cloud-native services.
2. Mixed-type workloads, for which hybrid models provide the most effective outcome; that is, the stable core of the system may be deployed on on-premises infrastructure, while testing, peak, backup, or analytical subsystems remain in the cloud environment.
3. Mature, stable, and resource-intensive workloads, characterized by consistently high cloud costs alongside limited utilization of the benefits of elasticity.

Table 1.

 Comparative characteristics of digital workload deployment options

Accordingly, based on Table 1, repatriation is most justified in the context of a specific workload profile; that is, if a system is characterized by high and stable levels of resource consumption, significant performance requirements, limited dependence on proprietary provider services, and strong requirements for control over the computing environment, migration to on-premises or colocation systems becomes economically and architecturally justified. If the workload is mixed, a hybrid model proves more appropriate, as it allows for the distribution of roles across infrastructure environments [11].

The practical significance of the obtained results is confirmed by the analysis of corporate case studies. In particular, the experience of Dropbox’s infrastructure transformation demonstrates that migrating core data storage and processing components to a proprietary platform may be associated with more precise cost management, enhanced control over the environment, and a redistribution of architectural control [3]. A similar pattern can be observed in the case of 37signals, where the reassessment of the cloud strategy was driven by the need to optimize long-term operational expenditures and improve the transparency of infrastructure economics [12].

It is also essential to substantiate the need for a comprehensive assessment of workload suitability for repatriation, which should take into account the full set of relevant characteristics, including the predictability of the consumption profile, the degree of dependence on cloud services, the scale of fixed costs, performance requirements, the need for data localization, and the level of organizational readiness to operate and maintain proprietary infrastructure. For a more illustrative representation, these parameters are presented graphically (Figure 2).

Figure 2. Key factors determining workload suitability for repatriation, compiled by the author

Based on the obtained results, it can be concluded that cloud workload repatriation is most effectively implemented within a selective scenario. Modern organizations are characterized by the distribution of functions across multiple deployment environments; therefore, particular value lies in a model in which the stable core of the system is migrated to a private or colocation environment, while variable, backup, or rapidly changing components remain in the public cloud. Such a configuration makes it possible to combine the advantages of long-term economic predictability and architectural control with the benefits widely described in the literature on cloud systems, in those areas where cloud infrastructure is genuinely justified and required. Accordingly, three scenarios can be distinguished (Figure 3):

Figure 3. Deployment options for corporate workloads depending on their characteristics, compiled by the author

Thus, cloud workload repatriation represents a tool for enhancing the alignment between technological architecture and the economics of a digital system, the effectiveness of which depends on the accuracy of workload profile diagnostics, the quality of long-term cost assessment, and the organization’s ability to establish a rational configuration for the deployment of digital resources in accordance with actual business requirements.

Conclusion. Thus, based on the obtained results, it can be concluded that decisions on cloud workload repatriation cannot be made without a comprehensive architectural and economic assessment, which involves analyzing the workload profile, the level of technological dependence, the scale of costs, performance requirements, and constraints related to data and regulation. At the same time, repatriation is most justified for mature, stable, and resource-intensive workloads, for which prolonged operation in the public cloud is associated with high costs alongside limited utilization of the benefits of elasticity and managed cloud services.

The practical significance of the study lies in the fact that the proposed methodological approach can be applied in auditing cloud expenditures, evaluating alternatives, designing hybrid deployment models, and shaping an organization’s long-term digital strategy. The scientific significance of the study consists in advancing the understanding of the dynamics of cloud deployment value and in refining the conditions under which migrating workloads from the public cloud contributes to improving the architectural and economic efficiency of the digital environment.

References:

1. Armbrust, M. A view of cloud computing / M. Armbrust, A. Fox, R. Griffith [et al.] // Communications of the ACM. – 2010. – Vol. 53, No. 4. – P. 50–58. – DOI 10.1145/1721654.1721672.
2. Asundi, J. Using economic considerations to choose among architecture design alternatives / J. Asundi, R. Kazman, M. H. Klein. – Pittsburgh : Software Engineering Institute, Carnegie Mellon University, 2001. – (CMU/SEI-2001-TR-035). – DOI 10.1184/R1/6585749.v1.
3. Dropbox Engineering. Inside the Magic Pocket [Electronic resource].  – 2016. – URL: official Dropbox Tech page. – https://dropbox.tech/infrastructure/inside-the-magic-pocket  Accessed: 21.03.2026.
4. Flexera. 2025 State of the Cloud Report [Electronic resource]. – 2025. – URL: official Flexera report page. – https://info.flexera.com/CM-REPORT-State-of-the-Cloud Accessed: 21.03.2026.
5. Ferry, N. Towards model-driven provisioning, deployment, monitoring, and adaptation of multi-cloud systems / N. Ferry, A. Rossini, F. Chauvel, B. Morin, A. Solberg // Proceedings of the 2013 IEEE Sixth International Conference on Cloud Computing. – 2013. – P. 887–894. – DOI 10.1109/CLOUD.2013.133.
6. Khajeh-Hosseini, A. Research challenges for enterprise cloud computing / A. Khajeh-Hosseini, I. Sommerville, I. Sriram // arXiv. – 2010. – arXiv:1001.3257.
7. Kazman, R. ATAM: method for architecture evaluation / R. Kazman, M. H. Klein, P. C. Clements. – Pittsburgh : Software Engineering Institute, Carnegie Mellon University, 2000. – 102 p. – (CMU/SEI-2000-TR-004).
8. Kratzke, N. Understanding cloud-native applications after 10 years of cloud computing: a systematic mapping study / N. Kratzke, P.-C. Quint // Journal of Systems and Software. – 2017. – Vol. 126. – P. 1–16. – DOI 10.1016/j.jss.2017.01.001.
9. Li, A. CloudCmp: comparing public cloud providers / A. Li, X. Yang, S. Kandula, M. Zhang // Proceedings of the 10th ACM SIGCOMM Conference on Internet Measurement. – New York : ACM, 2010. – P. 1–14. – DOI 10.1145/1879141.1879143.
10. Mell, P. The NIST definition of cloud computing / P. Mell, T. Grance. – Gaithersburg, MD : National Institute of Standards and Technology, 2011. – 7 p. – (NIST Special Publication 800-145).
11. Tomarchio, O. Cloud resource orchestration in the multi-cloud landscape: a systematic review of existing frameworks / O. Tomarchio, D. Calcaterra, G. Di Modica // Journal of Cloud Computing. – 2020. – Vol. 9, No. 1. – Art. 49. – P. 1–24. – DOI 10.1186/s13677-020-00194-7.
12. 37signals. Our cloud spend in 2022 [Electronic resource]. – 2023. – URL: official 37signals Dev Blog page. – https://dev.37signals.com/our-cloud-spend-in-2022/ Accessed: 20.03.2026.