Cloud pricing is often presented as a simple "pay for what you use" proposition, perhaps with volume discounts -- or penalties. But pricing can be as complex as for any on-premises licenses or network bandwidth service. Exhibit A: reservation-based pricing, such as Amazon Web Services' Reserved Instances.
If you expect consistent, heavy use, Reserved Instances can provide substantial savings over owning your hardware, as well as over using AWS's standard as-needed pricing.
But with Reserved Instances or its equivalents at other cloud providers, such as Microsoft's Enterprise Agreement for Azure, you're betting you'll actually use the instances you reserve because you'll pay for them whether or not you use them. If you forecast less consumption than you end up needing, you'll pay full retail prices for the excess -- an amount likely not covered in your budget forecast.
For Reserved Instances to make economic sense in practice, you have to be good at capacity planning. However, this isn't a skill among many IT organizations, and it can turn Reserved Instances ito a sucker's bet for many, where the house (AWS, in this case) comes out ahead.
But if you're good at capacity planning, you could save real money. In particular, large enterprises could easily save millions of dollars each year.
Still, there's a middle path for organizations that aren't so good at capacity planning or don't have predictable usage patterns: Purchase at least 50 percent of your instances as reserved to get a consistent discount for a large portion of what you know you will use. (When you purchase a Reserved Instance, the reservation is automatically applied to running instances that match your specified parameters.)
Over time, it should be easier to make accurate usage forecasts and safely increase the amount of Reserved Instances purchased. You'll have more history to go by, and to aid in your planning, usage-based accounting tools are beginning to support the notion of reserved cloud capacity. Now is a good time to explore such tools and get a handle on your usage patterns.