ECS - Services initial CPU consumption

I have an ECS cluster with some services. I'm having a hard time determining the Task CPU. These services use very low CPU while running. But when they launch they use an interesting amount of CPU. So I'm confused, since If I raise the Task CPU, I need more ec2 instances, and after the service starts, the CPU will remain almost unused and I'll be spending more money on infrastructure. But if I decrease the task CPU, the service will take a long time to start, I need to raise the health check interval but of course, infrastructure will be cheaper. So I'm not sure which is the best approach for this scenario