The chase is on to catch up to Amazon Web Services, a leading infrastructure as a service player, and Google has launched an all-out price war that Microsoft's Azure, Rackspace, and the rest of the pack will have to follow.
It's a infrastructure as a service (IaaS) pack that's getting increasingly crowded too. IBM's SoftLayer, Hewlett-Packard, Oracle, Verizon, and a bevy of others are all playing the IaaS game.
But Bernstein analyst Carlos Kirjner is betting that IaaS really boils down to a three-horse race. Amazon, Google, and Microsoft. The other players either have economic conflict of interests or won't be able to scale in the deflationary cycle that AWS and Google are about to start in technology infrastructure.
Kirjner was riffing on Google's recent price cuts across its Compute Engine server types. He said:
We believe the number of players who can compete at scale with AWS (i.e., beyond specialized niches) can be counted on the fingers of one hand. This is because playing in the big boys' IaaS league requires a set of assets, skills, and experiences that very few businesses have: They range from building distributed computing infrastructure including hundreds of thousands and in the not-so-distant future millions of servers, to the software development skills to create and support the literally thousands of different services and features available on that infrastructure to enterprise and government customers, to the experience base on hyper-scale distributed computing required to operate and evolve the underlying technology infrastructure with reliability and in a cost effective way.
It's hard to disagree with Kirjner that only a few companies can handle hyperscale infrastructure, but the field may be larger than he suggests. He noted IBM can't be dismissed, but doesn't have the experience or expertise — even with SoftLayer — to compete. Kirjner's key point is that IBM doesn't have the economic incentive to compete in IaaS. That final point may be on target since IBM — even with the move to ditch low-end servers to Lenovo — still sells a lot of hardware.
That economic argument can be made for many players in the IaaS space. Telecom giants — think Verizon — aren't going to get into price wars over IaaS, and will stay upstream. Oracle sees IaaS as a way to sell its cloud applications and platform. Infrastructure to Oracle is the price of admission to the cloud.
In our view, one should expect the IaaS market to have three, and in an extreme case four, large competitors, which in aggregate should account for 60% to 80% of the market. Many others, including Verizon, AT&T, VMWare, HP, Centurylink, BT, Cisco have entered the IaaS market (or at least have issued press releases saying they were entering), probably because they recognize the vast opportunity it presents, and can do enough arithmetic to decide that a fractional market share in a very large market may still be interesting. We have very little doubt that they do not have the right computer science and engineering skills and assets to compete with Amazon, Google, and Microsoft.
Economics frame much of Kirjner's argument. Amazon, Google, and Microsoft will have no choice but to cut prices about 20 percent to 30 percent a year, but cost of goods sold, server depreciation, Moore's Law and power consumption will also fall at the same clip. In other words, gross margins may stay stable. However, it's unclear that Amazon, Google, and Microsoft can make money on IaaS. He added:
We think the public cloud business works at scale because AWS and possibly Google Compute Engine will be able to operate at very high server utilization; high enough to drive gross profits on a per-server basis and at a large enough scale to offset the fixed costs (R&D and G&A).
Top line: Google and AWS may combine for about $40 billion in revenue by the end of the decade while cutting prices thousands of times.
So why wouldn't some other tech giant enter the fray? Google, Amazon, and Microsoft are likely to set off a deflationary IT cycle. Few enterprises will be able to match the utilization of the cloud's big three. As a result, they won't buy servers. Many other cloud players, who happen to sell hardware, are going to blink. They won't want to contribute to the unraveling of their key businesses. "One server or switch bought and used by AWS or Google at high utilization may meet the demand that would require two servers bought by enterprise customers," said Kirjner.