HYBRID MULTI‑CLOUD AI ORCHESTRATION USING SERVERLESS WORKFLOWS

Enterprises increasingly deploy AI workloads across heterogeneous cloud environments to meet demands for
scalability, compliance, and performance. This paper presents a hybrid multi‑cloud orchestration framework
that leverages serverless workflows to automate AI model training, validation, and deployment across AWS,
Azure, and on‑prem Kubernetes. We integrate federated learning for data‑sovereign training and enforce zero‑trust
security policies via a service mesh. A prototype demonstrates near‑linear scaling on GPU clusters, sub‑second
inference latency during autoscaling events, and robust policy enforcement with zero unauthorized access under
simulated attacks.