Which solution will meet these requirements?

Explanations:

Burstable performance instances in unlimited mode can handle sudden increases in demand by utilizing additional CPU credits, which is useful for irregular spikes in utilization. Scaling based on requests allows Elastic Beanstalk to respond dynamically to actual application load, aligning with the requirement for auto-scaling based on demand.

Compute optimized instances provide high CPU performance, but they are designed for consistently high CPU usage rather than infrequent, unpredictable spikes. While scaling based on requests is beneficial, compute optimized instances are not cost-effective for occasional spikes and could lead to unnecessary expenses.

Scaling on a schedule does not match the company’s requirements since the latency issues occur unpredictably. Scheduled scaling is beneficial when demand patterns are predictable, but it would not address sudden, unanticipated CPU spikes.

Predictive scaling uses historical data to anticipate future demand, but this is not well-suited for the company’s infrequent, irregular spikes in utilization. The latency issue does not occur on a regular, predictable schedule, making predictive scaling less effective in this scenario.