Cloud resources are invisible. Processing power, frameworks, messaging infrastructure, Web APIs, virtual machines, even storage. We don’t see them. Often we don’t pay for them, and when we do, we rarely pay much. How does that work?

Cloud resources from Amazon or Apple or Google or Microsoft or Oracle or SAP or whoever aren’t provisioned by magic. Behind those 1s and 0s, those click-to-approve license agreements, are huge data centers. Those data centers are often sited in remote locations near hydroelectric power plants—that is, cheap electricity. Size matters. These centers are filled with thousands of racks, tens of thousands of servers, most often inexpensive x86 servers running Unix or Linux or Windows.

A cloud host’s servers looks similar to those you’ll find in your company’s own data center—that is, if you still have a data center: little 1U- or 2U-high pizza boxes with one or two processors, lots of RAM, and a few spinning hard drives. The difference is that only rarely will you find a brand name like Dell or HP or IBM on the servers. The bulk of data center servers are anonymous and beige, built to the cloud provider’s specifics by the lowest bidder.

Given the quantity of servers—tens of thousands—and the incredible redundancy in the host infrastructure given the extensive use of virtualization, the individual server is quite unimportant. It’s not worth paying extra for quality servers.

Move beyond the servers, and within the cloud data centers you will find quality nearly everywhere else. The switches. The routers. The redundant power supplies. The backup generators. The control systems. Those are where you find name-brand investment, not in the essentially disposable servers, which is what the typical consumer (or IT administrator) cares about.

What matters, beyond low cost and redundancy, is density. Lots of processing power in each CPU. And lots of storage in each disk array. As far as I know, the greatest density in server storage is the 6TB 3.5-inch drive, such as the helium-filled Ultrastar recently introduced by West Digital’s HGST division. Pack a bunch of those in a pizza box, or in a blade server, and you’ve got some service capacity for the public cloud.

Someday, someone is going to have to pay for those data centers, including both their capital and operating costs. As the price of cloud computing keeps dropping, and as more and more services become available for free, it’s important to remember that these cloud data centers are expensive. Right now, many cloud services are offered at a loss. Advertising won’t always make up the difference.

We can be sanguine about using the cloud to reduce the cost of enterprise computing, and yes, the economies of scale suggest that cloud computing should generally be more cost-effective. But it can’t always be free. The race to the bottom on cloud pricing won’t last forever. Someday, we may have to pay real money for that hot air. That’s not something to look forward to.

Where do you see cloud pricing going? Write me at alan@camdenassociates.com.

Alan Zeichick, founding editor of SD Times, is principal analyst of Camden Associates.