It's always a delicate balance to find that sweet spot where your application is highly available to your audience and at the same time you're not paying for expensive servers that are underutilized. Let's consider Compute options from the perspective of cost savings and availability. Spot Instances. For Compute intensive application, Spot Instances are the absolute best choice. It simply means buying Amazon's unused EC2 compute capacity at a huge discount. One disadvantage is that if your application requires near 100% availability, then this may not be the optimal solution since Amazon can reclaim the Spot Instances from you on short notice. That said, they can be combined with other Compute options to meet the availability requirements at a higher cost. Serverless. If your application only runs short-lived transactions, serverless may be able to save you a lot in cost. 

However, if your applications run long processes or has a very high transaction load, say hundreds or even thousands of transactions per second, then serverless may not be the best choice. Auto Scaling. Auto Scaling not only helps with cost savings, but also helps with availability because it forces you to use more than one availability zone in your design in order to take full advantage of it. Reserve Instances. This is simply the action of pre-paying for your EC2 compute capacity if you know you'll be using it for several months or even years. This is guaranteed capacity, which in contrast with the Spot Instances, these are guaranteed to be available to you. It's a great cost saving options right in the middle between on-demand pricing and a spot instance pricing. As with everything in life, the actual answer to making your application highly available and also running it at a low cost lies somewhere between these options. For instance, combining auto scaling with spot pricing gives you higher availability while keeping costs down.