The decision rarely hinges on cost first — it hinges on what the data is. Classify each dataset by sensitivity and legal residency requirement before evaluating any platform. Highly classified or residency-bound data narrows your options immediately, and trying to reverse that decision after go-live means a migration under audit pressure.
On-prem gives you sovereignty but transfers patching, backup, and uptime onto your team. Cloud reduces that burden but introduces shared-responsibility boundaries that are easy to misread. We score each option on day-two operations — who patches, who restores, who is paged at 3am — because the launch is the cheap part and the next five years are where budgets are won or lost.
Most real deployments land in the middle: sensitive records stay on-premise or in a private cloud inside Thailand, while stateless compute and non-personal workloads burst to public cloud. The key is a clean boundary so that crossing it is an explicit, auditable event rather than an accident of configuration.