Yep, we could handle it easily on one machine, if we gave it unfettered access to all the resource, but our throughput varies during the time of day, so dedicating a machine to it to handle the peak throughput wouldn't make financial sense at 3am at night. So it's far simpler to scale pods with known resource usage as needed.
It also helps prevent issues when throughput suddenly doubles because of a business decision that you're left out of the loop on.
So autoscaling pods is ideal for our use case.
