By experience, an optimized MongoDB usually needs more IO and memory, so we can enable a horizontal scale during the live incident.
So a larger memory SSD would be preferred.
For AWS, eg. x2iedn or x8g(require to switch ARM) would be an option.