Seen this pattern many times -- the root cause is usually not the cloud itself but a combination of over-provisioned resources, missing cost visibility, and reactive budgeting. Things like untagged resources, idle NAT gateways, oversized instances, and no automated scale-down policies quietly eat the budget.

At UPSystems we typically start with a quick resource rightsizing + tagging audit, set up real-time cost alerts, and move heavy AI/LLM workloads to on-prem where it makes financial sense. Happy to share our FinOps checklist if useful -- feel free to DM.