The Limits to Cloud Price Reduction
Although public clouds benefit from economies of scale from massive and centralized data centers with high utilization, and continuous improvements in cost per unit of processing capacity from Moore’s law, they’re unlikely to be able to drop prices more than 15 percent annually over the long haul, barring some massive technology discontinuity such as free electric power or disruptive new computing paradigms. The simple reason is that even if CPUs were free, there would still be infrastructure capital expenditures, such as for data centers, and ongoing operating expenses, such as for power. These cost elements don’t follow Moore’s Law.
The last issue of IEEE Cloud Computing (download post-print version) features an article where we explain why Moore’s law does not apply to cloud data centers, and how a repeated cutting of prices to follow Moore’s law would considerably reduce cloud provider profit margins. Since cloud computing became widely available, in spite of well-publicized price reductions from a variety of providers, Amazon public cloud prices have only fallen at 10.5 percent annually. Google, although it promised to follow Moore's Law principles for cloud pricing, has dropped 20.8 percent annually in the last 3 years. We show how public clouds, despite their economies of scale, are unlikely to be able to drop prices more than 15 percent annually over the long haul without impacting their margins and/or bottom line. In other words, the price of a unit of computing power in the cloud is likely to drop roughly 50 percent every five years, at best.
Similar logic regarding cost structure applies, in the best case, to private clouds. Of course, this requires several things: 1) that a private cloud can achieve statistical and other economies of scale, 2) that it can leverage best practices and open standards such as Open Compute server designs and OpenStack and OpenNebula, 3) that the firm can procure equipment or even full data centers at competitive prices through volume discounts. In this case, their cost for unit of compute should also be able to decline at more or less the same rate as public cloud providers.
Acknowledgement: We would like to thank Joe Weinman for his highly valuable feedback and recommendations during the preparation of the manuscript.
