Abstract: Finding the right balance between cloud costs and application performance has become much more challenging due to the quick adoption of multi-cloud architectures. The majority of autoscaling solutions are rule-driven, reactive, and intended for single-cloud scenarios. This restricts their applicability in intricate multi-cloud settings with disparate pricing schemes, data transfer fees, provisioning delays, and spot/preemptible instances. In order to reduce costs while maintaining service-level objective (SLO) limitations, this article presents MCCAS (Multi-Cloud Cost-Aware Autoscaler), an AI-driven autoscaling platform that intelligently optimizes resource allocation across several cloud providers. Three interrelated parts make up MCCAS: (i) a workload forecasting module that uses deep learning to predict short-term demand and uncertainty; (ii) a thorough cost and risk modelling layer that takes provider-specific pricing, migration overheads, and pre-emption risks into account; and (iii) a hierarchical reinforcement learning (HRL) decision engine that makes a distinction between short-term tactical scaling actions and long-term strategic placement decisions. MCCAS uses hierarchical control to make the decision space simpler. This makes it simple to adapt to changes in pricing and workload. With rewards that explicitly penalize expenses, SLO violations, and inefficient migrations, the suggested method sees autoscaling as a sequential decision-making issue with limitations. Using real workload traces and simulated multi-cloud pricing scenarios, a comprehensive experimental design is shown along with comparisons to rule-based, prediction-only, and single-cloud learning-based autoscalers. The outcomes should show that cost-conscious, AI-driven multi-cloud autoscaling may drastically save operating costs without sacrificing application performance guarantees. This study offers researchers, cloud architects, and FinOps teams a practical and repeatable method for implementing intelligent autoscaling solutions in actual multi-cloud systems.

Keywords: AI-powered autoscaling, resource provisioning, cloud economics, workload forecasting, finOps, multi-cloud computing, cloud cost optimization, reinforcement learning, and hierarchical reinforcement learning.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14401