Cloud Computing Network Effects, Or Why Tim O’Reilly and Nick Carr Are Both Wrong
Posted by Bob Warfield on October 28, 2008
Oh boy, picking a fight with Tim O’Reilly and Nick Carr in one blog post? Have I lost my mind? Is this just pure flame baiting?
No, there is a method to my madness, and something to be learned by dissecting this back and forth between Tim and Nick.
First, what is the back and forth? It all started with a post by Tim O’Reilly that presented some innovative thinking about cloud computing. Tim has been thinking hard about the role of network effects in bringing an unfair advantage to web companies, and in this post, he wonders aloud whether there are any network effects for Cloud Computing. This post was promted by Hugh Macleod creating some buzz around the idea that Cloud Computing would devolve into a huge monopoly worth potentially multiple trillions of dollars.
MacLeod’s post was interesting, but pretty bare of strong theory for why the multi-trillion dollar monopoly scenario should come into being. How could a single company come to so dominate The Cloud?
O’Reilly argues in his post that it is network effects that give web companies unfair advantage and that there are none for Cloud Computing would-be monopolists. At best Cloud Computing is a commodity and commodities often have economies of scale (e.g. they can buy equipment or power in enough volume to get it more cheaply than others), but commodities do not have network effects, so there is no outsize economic leverage for the winners. I disagree with Tim on this, and will explain why in a moment, but we need to plunge ahead in the recitation of the Tim/Nick argument further before I do that.
BTW, a “network effect” is simply a situation where each incremental new user is worth more than the last one, which drives ever increasing advantage for those who learn how to grow more quickly than others. Eventually it creates an unfair advantage so large that it cannot be overcome by competition and gives the owner unfairly strong profitability.
Let’s now turn to our other contenda, Nick Carr, for his argument about why Tim is wrong. Unfortunately, this one must be short, because Nick doesn’t seem to understand the term “network effects” well enough to hold up his end. He argues that simply because the links used in Page Ranking (Google will weight search results more highly if lots of pages link to a particular result page) are available to any search engine, Google gains no network effect advantage in utilizing them.
So, we’re back to Tim, who responded both in a comment to Carr’s column and in his own rebuttal post. O’Reilly lists no less than three ways in which Google benefits from network effects that Nick Carr either missed entirely or got wrong. First is that Google is simply better at spidering the web’s network of links than their competitors. Whether its that they have simply indexed more pages, or seen more web behavior by achieving a big early lead and relentlessly exploiting that advantage to tune their algorithms (if I have seen further it is because I have stood on the shoulders of other men, but I digress with this Isaac Newton related quote), Google seems to do it better, or at least well enough that nobody can gain enough temporary advantage to grow relative to them.
Carr says this is simply a matter of Google having built a more powerful indexing engine, but O’Reilly is right when he says the reason Google did that is to gain the advantage of more network effects. Point to O’Reilly.
Second network effect Tim brings up is the Page Rank algorithm itself. While Carr is right that this is available today for anyone to exploit, Google saw it first and built a huge lead before everyone else figured it out. I believe they are far more secretive today about what they’ve figured out and that they’re further extending their lead by virtue of network effects. Second point: O’Reilly.
Last network effect example is Google’s advertising auction. Auctions are well known to benefit from network effects. This is why eBay so completely owns the online auction world. Think about how it works. A buyer is more likely to buy if they can find what they want, so they go where there are more sellers. A seller is more likely to list if there are more buyers. This ensures both a higher likelihood of selling at all, and a higher price with more buyers to bid against each other. More buyers leads to more sellers, which leads to more buyers, and so on. It’s a beautiful thing and Google applied it to their advertising. As Tim so effectively points out:
It isn’t that the advertising-side network effect has anything to do with Google’s dominance of search, but rather, that Google’s dominance of search is central to the design of their ad auction. You see, while Yahoo! (nee Overture) sold keyword advertising to the highest bidder, Google realized that they could mine their users’ clickstream activity to predict which ads would be most likely to be clicked on, and by what ratio, and thus sell to the best combination of price and actual click through. Thus: higher revenue, more ability to invest in infrastructure, better results for advertisers and users, thus more users, thus better data, thus better results for both organic search and advertising (both of which do, in fact, matter to users, no matter what Nick thinks).
So Tim convincingly won that bout, and I’ve seen no response from Carr or others that does much to refute what he is saying. It’s only been a day, so perhaps Nick is formulating another response, or maybe he has thrown in the towel.
But Why is Tim O’Reilly Wrong to Say There Are No Cloud Computing Network Effects?
Aha! This is an important point to discuss now that we have dealt with the historical back and forth. There are essentially four points in Tim’s rebuttal post to consider. The first three are about types of Cloud Computing, and the fourth is about some called the “Law of Conservation of Attractive Profits”.
Network Effects that Benefit Utility Computing
Start with Utility Computing, which is Tim’s first kind of Cloud. Amazon Web Services is the poster child. Tim says no later users benefit from the growth of Amazon other than perhaps through a rise in Amazon’s commitment to the business. He admits a few edge cases. Maybe there will be more developers skilled with AWS so it’ll be easier to build AWS apps. That is contra-indicated by the simplicity of the AWS API’s. To gain lock-in, Amazon must increase API complexity, but there is too much momentum towards open source so its hard.
But wait Tim, there are some interesting network effects already built into the Amazon Cloud, whether or not we see folks taking advantage yet or not. Here is a key effect: it costs you very little to move data around inside the Cloud compared to what it costs you to move data in and out of the Amazon Cloud. Hence, if applications need to exchange data, it is advantageous to do so by having the applications exist inside the same cloud. And, the more data that goes into that cloud, the more advantageous it will be to move more apps into the cloud. Just like the auction scenario. Are those benefits as strong? Maybe yes, maybe no, we have yet to see.
Here is another key network effect related to elasticity. Elasticity is a property of Cloud Computing whereby I can rapidly provision a new server to meet the demands of scaling. I can get a new server on Amazon in something like 10 minutes, for example. I pay for it by the hour, and when I’m done, I return it to the pool.
What does that have to do with network effects? Well, consider a small cloud. If the cloud is small in relationship to its largest user, there is very little room for elasticity. Elasticity represents unused capacity that the Cloud Vendor had to pay for in order to have inventory to sell when demand comes. The larger the cloud is relative to the elasticity needs of a particular customer, the cheaper it is for that vendor to provide the elasticity because the extra inventory is a smaller fraction of overall inventory. In fact, if you are a large organization at all, you should want to know the details of how much elasticity headroom there is relative to your needs.
Network Effects that Benefit Platforms as a Service
OK, that’s two examples of network effects that accrue to larger vendors of utility computing. Let’s move on the the Platform as a Service type clouds. PaaS platforms hide machine instances behind higher level API’s. A pure PaaS platform will have advantages similar to a Utility Computing platform in terms of network effects, but there is an opportunity for a further advantage if you have what I call an “Affinity Platform.”
Let me give two examples: Force.com and Intuit’s QuickBase platform. Do you see what I mean by an Affinity Platform? These are platforms that started from an application. Force.com started from Salesforce.com. QuickBase started from Quicken/QuickBooks. There are many others including the API’s and platforms for things like Facebook.
Clearly Affinity Platforms have network effects and lock-ins. Each additional person using an application brings more value to the platform associated with the application. This value accrues for everything from rapidly reaching that installed base (ala App Exchange) to being able to perform interesting integrations, data aggregation, and other ways of adding value across multiple tenants, applications, and vendors. It’s an ecosystem, in short, and is very much amenable to network effects.
Cloud Based End User Applications
This definition refers to individual web applications that were formerly delivered on a PC such as spreadsheets, word processors, and the like.
Why has Microsoft Office maintained its dominance for so long? Is it because it was a commodity and they just undercut everyone on price to create a low margin business. No, not at all. O’Reilly suggests companies like Rackspace show what low margin commodity businesses look like, not Microsoft.
Microsoft created a simple network effect in their Office Suite with integration, and in the ubiquity of being able to exchange data with the largest number of users. I might like a particular spreadsheet better than Excel (says he who created Quattro Pro long ago), but I don’t like it enough better to forgo an integrated Word Processor. After a while, even those who did some other spreadsheet well enough to use an alternate word prcoessor can no longer get by because they can’t exchange data effectively with those who use the “standard” MIcrosoft Office applications. Yes, there are ways around this, but they simply aren’t convincing.
Where I have been completely disappointed by the current crop of Microsoft Office-as-web-app wanna be takover artists (like Google Apps, Zoho, et al) is that they are just a subset of MS Office running on the web with decent but not great data compatibility. And I’m gonna switch for that? Please!
They are so missing the boat by not focusing on what the web enables them to do far better than any desktop application: collaboration. Collaboration, by its definition, would make for an effective network effect for any Cloud Based End User application. Why don’t these vendors get on with discovering and delivering effective collaboration? There are tons of specific uses of these tools, especially for spreadsheets, but also for word processing, that cry out for this. Just look at how painful budgeting in Excel continues to be for most organizations. Roll up a sales forecast using spreadsheets, or do any one of a number of similar exercises. You see my point.
The Law of Conservation of Attractive Profits
I love it when guys make stuff up and call it a “Law” so that its harder to argue with and makes them sound more smart. This one is attributed by Tim O’Reilly to Clayton Christensen. It basically says that when a part of the market commoditizes, there will usually be an opportunity to recapture profits lost to commoditization in an adjacent stage of the value chain. In other words, while PC hardware commoditized and was not a profitable business for many, Microsoft got the adjacent OS space and was hugely profitable.
O’Reilly’s take away is that Cloud Computing is real, but that it is a commodity, so find that adjacent stage in the value chain and stake out an unfair advantage using network effects. In general, O’Reilly thinks those network effects accrue from data more than software. It’s how the software is used, in other words.
I won’t disagree with Tim on this last point, but I think I have shown how there can be powerful network effects for the three models of Cloud Computing he has talked about. Perhaps elegantly for Tim, my ideas have mostly revolved around how the software in each case is used–design software that benefits from being able to communicate with other software in the same (largest) cloud, design software that benefits from users that have an affinity through using the same application (like Salesforce or QuickBooks), and design desktop app replacements that benefit from people who want to use it to collaborate, rather than just use it to do what they used to do on the desktop.
Any thoughts on this Tim? Nick?