A company has an on-premises web application that experiences latency spikes twice monthly. At the start of a spike, CPU utilization immediately increases to 10 times normal. The company will migrate the application to AWS and wants automatic scaling to handle increased demand. The company will use AWS Elastic Beanstalk for deployment. Which solution meets these requirements?

The correct solution leverages burstable performance instances in unlimited mode and scales based on requests. Burstable performance instances (like T-series) are cost-effective for applications with fluctuating CPU usage, and unlimited mode allows them to burst above their baseline for extended periods when needed, which is ideal for sudden, high CPU spikes. Scaling based on requests directly addresses increased demand on the web application. Compute optimized instances are designed for sustained high CPU usage, which is not the primary characteristic described (spikes, not constant high load), making them less cost-effective for this scenario. Scaling on a schedule is unsuitable because the spikes are not predictable enough to be scheduled accurately. Scaling on predictive metrics is also less effective here because the problem describes an immediate, 10x increase in CPU, which is often reactive rather than proactively predictable.