The correct answer is 'Pay-as-you-go compute instances.' Since the application demands high computational power during active periods and none during inactive periods, a pay-as-you-go model allows the company to provision high-performance computing instances only when needed, optimizing costs by not paying for idle resources. This is in line with the benefits of Infrastructure as a Service (IaaS), where resources can be dynamically allocated based on demand. 'Reserved compute instances' would not be ideal as they assume a constant need for resources and would not account for the company's intermittent usage pattern. 'Dedicated physical servers' would be inefficient and costly for their sporadic usage requirement. 'Subscription-based virtual desktops' are designed for end-user environments and are not suitable for CPU-intensive computing tasks.