A small server can run a serious data platform, but only if there is room for the jobs as well as the services that wait for them.
The first measurement
The original plan kept the orchestration and deployment tools together on a small node. The team measured how much memory they used rather than relying on estimates. That gave the platform a starting budget, with a condition for moving to a larger machine if the planned work could no longer fit.
The decision changed
Later, the projected requests for the next stage exceeded the budget. The team chose a larger node before adding the orchestration service. The trade-off was a higher monthly cost in exchange for keeping room for the planned stack and its jobs.
What this taught me
A resource limit is useful when it prompts a decision, not when it forces the system to lose capabilities it needs. The measurements and their assumptions remain part of the record.