The Inventor’s Dilemma
This post is an amalgamation of many of my prior Cisco posts, trends in networking and Clayton Christensen’s book. I started a thought process in my Notebook 11.02.11 post and this is extension of that thinking. I wrote on 11.02.11:
“Datacenter Switches and Mobile Devices: I was sitting through a presentation from a merchant silicon team last week. They were going over their designs in detail as well as roadmaps and application examples. They put up a slide matching chips to switch reference designs. A customer could choose chip A and build 48x10G switch or chip B with 48x10G and 4x40G design, etc. I was looking at the slide and thinking where have I seen this before and then it occurred to me: the mobile device market. One of the reasons I have been negative on the mobile device market is it really down to four operating systems: IOS, Android, Windows and RIMM. Apple is a fully integrated (vertical) and closed system. That is the most valuable franchise in mobile devices. RIMM has the oldest OS that was designed to solve problems from ten years ago. If you want to design a smartphone today, the process is to pick an OS: Android or Windows. Then you get a reference design from QCOM or Mediatek, etc. The real value is in the user interface (UI). I know marketing, manufacturing, distribution all matter too, but the mobile device market is bifurcated into Apple (the closed OS with ASICs) versus the Windows or Android OS phones with developer built UIs and merchant silicon (reference design) mobile device companies. I am crossing analogies to explore a thought path. Will the switching market go the way of the mobile device market? Will CSCO someday have to make a choice like Nokia or will they go the way of RIMM? Will a new company or legacy company quickly rise to take share because they built a vertically integrated switching solution like the iPhone? Will the rise of merchant silicon and reference designs create a vast market of white boxes driving down price points in which the differentiation and value is in the software, i.e. the customer UI? Just a thought path I am exploring.”
Here are six network diagrams. Take a look at the graphic and see if you can guess the year of each diagram. If you guessed reading top left across to bottom right: 1991-1995-1999-2003-2008-2011 you are correct. I consolidated these diagrams from current websites as well as the Internet Archive. No real mystery in the diagrams. It does not require an engineer to realize that we have been replicating the same network design for twenty years. When I first posted a theory about Moore’s Law Exhaustion, which now seems dated as I wrote it in July, I was specifically referring to the replication of this network architecture in product cycles based on the next iteration of silicon, form factor and densities.
Cisco reports a large installed base for the Nexus platform. At their recent technology field day the company presented the Nexus customer base as “19,000 NX-OS customers, 50K Nexus 5000 chassis, 18,000 Nexus 7000 chassis and 3 million 10 gigabit Ethernet ports shipped!” The chart to the left shows Cisco’s quarterly switching revenue going back to 1999. Long time readers are familiar with my theory on the catch-up spend coming off the 2008 credit crisis. What is interesting to ponder about this chart is the direction over the next two years after the catch up spend was completed. It does show a topping out pattern and this might be reflective of product cycles, macro weakness or a changing market. When CSCO reports this week it will be interesting to see the direction of line. I drew in three dotted line variants.
In the world of switching it is relevant to understand where profit margins are located in the network diagrams show above. The exact profit margin varies vendor to vendor, but in the Leaf/Spine or ToR/Switch or ToR/Fabric designs (different terms for the same design), the calculations I have done show that profits can be broken down into two groups. Approximately 28-45% of the profits are in the Leaf/Spine or ToR/Switch or ToR/Fabric portion of the solution. The remaining 55-72% of the profits is in the pluggable optics – which the system vendors are reselling. Another way to look at would be to say that ~60% of the solution cost is the pluggable optics, which is marked up from the manufacture. I think this is a sustainable model – until someone decides to do something different.
That leads to the questions of what would be something different in networking/switching and who would do it? I have heard a number of people tell me that their strategy is to use multi-sourced commodity silicon (i.e. merchant silicon) and write their own code (i.e. in house switching code or some variation of Openflow). The quasi-vertically integrated vendors like CSCO, JNPR, BRCD, DELL, HPQ, do have a similar approach. Almost all the incumbent suppliers in this space are using chips from Fulcrum or Broadcom. Some suppliers use ASICs on the large switch platforms like the Nexus 7K. Merchant silicon has the potential to disrupt price in the market, because value can be pushed to the edges of the model. Using the handset market as an analogy, AAPL has a tightly integrated model (i.e. closed ecosystem) around IOS, iTunes and A4-A5 processor and their devices. A lot of value (i.e. profit) is created in this integration because the user experience is powerful.
The competitors to AAPL who use Android or Windows (i.e. merchant operating system (OS)) do not have as powerful a user experience and brand profile, but they can sure produce a powerful hand held device a lower cost point than Apple. That has an affect on the market especially when large legacy companies with dominant market share (e.g. Nokia had 40% mobile device share in 2008) are faced with lower cost devices coming from merchant OS/merchant silicon sources and a superior high end solution coming from Apple. This type of market structure could develop in the world of switching in which there are suppliers willing to live on much lower margins such as DELL, who use commodity silicon and a commodity OS. One or two other vendors might emerge who build a tightly integrated solution or maybe take a different architecture approach and they can demand a premium because they are solving problems that the legacy network design cannot solve such as scale, bi-sectional bandwidth and simplicity. All of the above brings us to the inventor’s dilemma. I think CSCO has three options to play or some combination of the three.
Plan A: Do nothing, keep the status quo, assume that 75% market share will forestall changes in the market structure and what has worked over the last twenty years will work for another twenty years. CSCO sells ~$3B a quarter in switching gear, has 10s of thousands of customers as cited above, armies of loyal and highly trained people around the world and plan A would be to continue to extend the twenty year run for another decade. Assume the worse case is someone does disrupt the profit model and in that case Cisco just rides it out without matching pricing over the decade and maybe Cisco has to buy a company if the pain gets to be too much. Cisco chooses this plan because disrupting the ecosystem model with armies of VARs and distributors and customer relationships is not worth the effort.
Plan B: Keep the twenty year model, but fight competition and change with price. Over time profit margins erode, but this will be a long process and we can always do something different when we are forced to act differently.
Plan C: Self Cannibalization. This is the most disruptive plan. This requires the CEO taking a team of people and putting them in a separate facility and telling them to come up with a new solution that obsoletes our existing installed base. Think IBM and Don Estridge inventing the IBM PC inBoca Raton, 1300 miles from the mainframe developers. Why wait for a competitor to cannibalize our installed base when we can do it ourselves. Plan C is the control our own destiny plan. We ride out the twenty year run and when it looks like the switching franchise might be threatened, we unveil a new solution to the market and change the playing field. Plan C comes with all sorts of risks and most of them are related to internal politics and inequality not to mention business and financial planning. Those are mainly the reasons why companies choose not to obsolete their own installed base and thus when faced with the inventor’s dilemma choose the least disruptive option to the business.
CSCO reports on Wednesday and I have no idea what to expect. Since the stock peaked in 2010, earnings events have been a disappointment with the exception of the August 2011 report. I heard one of the CNBC investment experts around noon recommending CSCO as a long and he might be correct, but I also think he has zero knowledge about CSCO’s business and the shifting sources of influence in networking. As usual this is just a blog and I could be entirely wrong about the various suppositions contained in the post.
* It is all about the network stupid, because it is all about compute. *
** Comments are always welcome in the comments section or in private. Just hover over the Gravatar image and click for email. **