What load balancer are you using? Does it have sticky sessions enabled by any chance?
Some insights - https://aws.amazon.com/premiumsupport/knowledge-center/elb-fix-unequal-traffic-routing/
Sample Tutorial just for you to double check the configs - https://docs.aws.amazon.com/autoscaling/ec2/userguide/tutorial-ec2-auto-scaling-load-balancer.html
Round Robin means that 2 targets on the ALB will each get every other request. If you want the majority of the requests to go to the new instance (until they're evened out) then you should use Least Outstanding Requests: https://aws.amazon.com/about-aws/whats-new/2019/11/application-load-balancer-now-supports-least-outstanding-requests-algorithm-for-load-balancing-requests/
However, you should also look into why the first instance is crashing when the CPU is only around 40%. See if there's a different resource other than CPU that's your bottleneck (memory, disk, network, artificial application limits, etc). If might be that you can get much better performance by changing instance families to one better suited for your resource requirements
Relevant questions
aws auto scaling routing
asked 20 days agoEC2 Auto Scaling
asked 7 months agoAuto Scaling Group not scaling based on ECS desired task count
asked 2 months agoaws iot auto scaling
asked a month agoAuto Scaling Group stuck in "Updating Capacity"
asked 2 years agoAuto-scaling is not working with Neptune cluster except when primary writer instance type db.r5d.xlarge.
asked a month agoAuto scaling question
Accepted Answerasked 3 years agoAuto Scaling Group SNS Notifications not being sent
asked 3 months agoIs it possible to use Auto Scaling for SAP Applications?
Accepted Answerasked 2 months agoAmazon EC2 Auto Scaling based on memory utilization
asked 5 months ago
Using Application load balancer. Stickiness is Disabled in Target group attributes