Handicapping the Cloud Race

The top 3 players offering general-purpose cloud computing services to the public are, in order of revenue: Amazon (Amazon Web Services), Microsoft (Azure), and Google (Google Cloud Platform). “General purpose” services are common to all or most apps and include compute, storage, networking, and monitoring. All of these vendors also offer a smattering of more special purpose services such as Hadoop. IBM has greater general-purpose cloud revenue than Google but lower public cloud revenue because a larger fraction is private cloud. Salesforce also has significant cloud revenue but that is special-purpose public cloud – primarily CRM.

I’m focusing on general-purpose public cloud because that category has the greatest market traction at this point. That said, I believe the cloud tsunami still has multiple inflection points yet-to-come. Nonetheless, I expect public cloud (whether general or special) to drive innovation more than private because it’s a larger market with a more diverse spectrum of users and uses.

Amazon has more cloud revenue than the next several players combined. They invented the general-purpose public cloud category in 2006 and, to their credit, have been innovating at a rapid pace ever since. Beyond externally-visible innovations in service offerings, under the hood Amazon has made significant innovations in virtual machine design, software-defined networks, and even formal software verification.

One challenge for Amazon is to avoid contracting a bad case of “feature-itis”. Currently Amazon has over 30 service offerings – more than Microsoft or Google. Navigating this maze can be daunting for potential buyers. Adding ever-more service offerings may actually increase time-to-market for a customer’s cloud-based system and reduce customer uptake. There is a tension here – and therefore a business opportunity. More on this in a later post.

Although being the most recent entrant (2010), Microsoft Azure has grown rapidly by taking advantage of being the obvious path into the cloud for the hundreds of millions of Microsoft product users. Microsoft is also taking advantage of its entrenched direct sales force, although some rumors of accounting shell games have surfaced. To their credit, Microsoft is thinking outside the Windows box and offering services based on Linux and other open source projects.

Microsoft faces two challenges: To not get trapped in a Windows-centric worldview that inhibits their adoption of open source alternatives (where significant innovation is occurring); and to not succumb to the fear that cloud sales will cannibalize product sales (it will and there’s nothing they can do about it). So far they are doing reasonably well on the former.

Google was second to arrive at the party in 2008. Unfortunately their initial offering, AppEngine, was very restrictive – it supported only one programming language (later expanded) and required the use of specific API’s for storage and network access. This restricted the available market to those writing new code or those willing to pay the cost of modifying existing code. Compare this to Amazon’s approach which supported any existing Linux stack un-modified (later expanded to Windows). To their credit, starting in 2012 Google began generalizing their offerings.

Compared to Amazon and Microsoft, Google’s challenge is more cultural in nature. To date, Google has been building infrastructure systems primarily for itself. AppEngine is a reflection of this – it is comprised of services originally built for internal use. To be a leader in public cloud, Google will need to evolve beyond this and market and sell directly to a broad spectrum of paying customers. Perhaps Google realizes this: They recently consolidated all of their general-purpose and special-purpose cloud offerings into a common group and appointed Diane Greene, former CEO of VMware, to run it.

As it stands right now, all of the Big 3 have comparable offerings and Microsoft and Google are providing stiff competition to Amazon not only in pricing but also in functionality such as by-minute pricing (vs. Amazon’s by-hour) and customizable compute instances. The intense competition between the Big 3 will lower costs and sustain the rate of innovation which will serve to expand the cloud even further.

Underneath any cloud are physical buildings stuffed with physical hardware, and the Big 3 all build, own, and operate their own datacenters. Datacenter capex, and especially opex, are significant components in cloud service economics and therefore a rich source of competitive advantage. Smaller players who do not control the complete hardware infrastructure inside the datacenter will be at a disadvantage.

The market footprint of cloud vendors is reinforced by network effects and some degree of lock-in. If a cloud customer has services X and Y running on one vendor’s cloud, and they then develop service Z which interacts with X and Y, it is very unlikely they will run Z with a different cloud vendor. Furthermore, as a customer runs more of their services with a particular cloud vendor, their switching costs to a different vendor increases.

It should therefore come as no surprise that revenue growth for each of the Big 3 is higher than the market as a whole which means revenue is concentrating with the leaders.

This capital concentration may open another front in the war: Silicon. All cloud vendors buy the same silicon (CPU’s, ethernet controllers, etc.) and are therefore un-differentiated at this level. The Big 3 have already been innovating at the board level and there is clearly opportunity for innovation at the silicon level. Such innovation will need to contend with the economics of silicon development which are quite different than software development. The costs of designing and fabricating a chip must be amortized over all units manufactured – if the volume is low this amortization cost can be prohibitive.

An interesting opportunity for new entrants is to target niche markets. An example of this is Nervana which offers a cloud-based deep learning platform into which users can plug their models. It’s interesting that Nervana’s cloud includes custom hardware (probably FPGA-based).

Finally, it’s not a competition of technologies or even of individual companies – it’s a competition of ecosystems. Nurturing a vibrant ecosystem on their cloud platform will be critical to any vendor’s long term success. Having many different users, running many different “experiments”, will create network effects and increase the probability that some of those experiments will lead to market-valuable discoveries – new paradigms, new services, new pricing models, etc.

I believe we are still in the first phase of the cloud tsunami and it’s clear to me that inflection points are in our future – in technology, service offerings, and pricing/business models. More on this in a later post…