Following up on a long running discussion between the likes of Hugh Macleod, Tim O’Reilly, and Nick Carr, Tim Bray today penned a piece entitled “The Shape of the Cloud.” In it, he argues that Amazon’s Web Services platform appears to him as not dissimilar from Altavista circa 1996. I don’t personally subscribe to this belief, but I could probably build a case to defend it. What I can’t defend, however, is his explanation for this assertion. In spite of my respect for Tim, I fundamentally disagree on several levels.
To look at why, permit me to indulge in a (significantly less snarky and massively less amusing) Fire Joe Morgan-style treatment of his piece. With all due apologies to those worthy masters, of course. For those of you unfamiliar with Fire Joe Morgan, Tim’s words are in bold, mine are not.
The Shape of the Cloud
There’s an interesting argument going on about the business-structure futures of the Big Cloud that everyone assumes is in our future. Some links in the chain: Hugh Macleod, Tim O’Reilly, Nick Carr, and Tim again. A few points about this seem obvious; among other things, Amazon Web Services is reminding me powerfully of Altavista.
Here are a few things that I think are true; the last section gets back to AWS and Altavista.
Monopolies Don’t Require Lock-in
Google has (effectively) a monopoly on consumer search. They have no lock-in; anyone who wants to can switch their default search provider in a few seconds. One could write a book, and several people have, about how they maintain their grip on the market, but let’s skip that and just see this as an existence proof.
With you so far. I accept both that Google has, effectively, a monopoly on search, and further that said monopoly does not depend on traditional IT lockin. Which makes it, in my opinion, an exceedingly unique market. I can’t think of another like it, and if pressed, for those somewhat similar, I don’t think cloud would be in my Top 5.
Which is to say, history tells us that Hugh Macleod’s vision of a single insanely-huge cloud provider is perfectly believable.
Whoa. There are a couple of assertions in there; let’s tease them apart. First, that search and cloud are conflated (without much discussion) such that the history of the search market is directly applicable to the cloud market. Sufficiently so that it’s possible to predict one based on the history of the other. Second, that Hugh Macleod’s argument that the cloud could produce “the largest company to have ever existed.”
Personally, I cannot agree with either argument. Taking them in order, the argument that search is predictive of cloud is wrong to me because:
- Search infrastructures are built upon the cloud, but are not bought, sold or consumed (typically) like the cloud. The user audiences are distinct, the barriers to entry and exit are distinct (for both customers and providers), the revenue models are distinct, and so on. Apart from the fact that the cloud is generally the infrastructure for the search, I see few commonalities between the technologies. One is infrastructure, one is an application: apples to oranges.
As for Hugh’s argument, I believe that to be unsupportable because:
- The combined revenue from Dell, HP, IBM, Intel, and Microsoft would not be enough to secure even the title of the largest business right now, let alone in history.
- We know Microsoft will be in this market. We know Google will be in this market. And, I’d argue, we know Amazon will be in this market. It’s also likely that some combination of IBM, Rackspace, Sun and others will be players in this market. Irrespective of whether or not we achieve some level of standardization, it’s not reasonable – at present – to assert that we’ll see any one of these players become ascendant and dominant in the way that Google or Microsoft are, because, fundamentally, search and cloud are different. I could easily migrate from Google tomorrow; migrating my application workloads from one provider to another, on the other hand, is non-trivial. Yes, even with Java.
Ergo, projecting one massive, monopolistic player in this market – as we see in search – seems at present to be unreaslistic.
To me, the monopoly of Google search demonstrates little with respect to the prospects for a cloud monopoly.
Low Barriers to Entry Win
· We should have learned this by now. You don’t have to look back very hard at recent decades to see examples of technologies which have become pervasive even though, when they started catching on, they weren’t nearly as good as the competition. The reason is, they were easy to learn and easy to deploy. I’m thinking, just for example, of Linux and PHP and HTML.
My sense is that the effortless-deployment threshold for the Cloud is somewhere below the effort required for AWS’s EC2/S3 combo; perhaps something like what a smart modern PHP-hosting service provides.
This section, I generally agree with, although I think it assumes (perhaps inaccurately) that Amazon will be unable to evolve in the design/development/deployment area. It’s also worth noting that – the unique-to-me search application market aside – lower barriers to entry generally result in more competition rather than less, making the case for monopoly more problematic.
Economies of Scale
· I don’t think they’re that big a deal. To play in this game, you’re going to need a few tens of thousands of servers, bulk-bandwidth deals with key carriers, and multiple data centers on the ground in North America, Europe, and Asia. That’s really expensive to build. But once a provider has got past the basic threshold, I’m unconvinced that their cost per unit of service is going to drop much as they get bigger and bigger.
My take-away is that the big Cloud operator doesn’t necessarily need to be a Microsoft or a Google or an IBM. There are multiple pathways by which an unheralded startup could end up on top of the heap.
Disagree, as might have been predicted by this piece. First, I think the above ignores the human economies of scale, in that Google et al can be expected to hire better, more qualified resources at a lower rate than your average business. Especially in the cloud, as the large providers can simply present more interesting problems to the best and brightest by offering larger datasets to operate from, larger and more unique infrastructures, and (particularly in this economy) greater certainty and predictability in terms of compensation.
As Nick Carr writes, as the marginal cost of the data decreases, its value increases: this is true in people, just as it is in hardware, storage, networking, capacity planning and so on. And, importantly, it’s more true for Amazon or Google than it is for, say, your local cloud startup.
I know a few vendors that simply cannot procure the scale-out talent that they’re looking for, not because they can’t pay, but because they can’t offer the types of challenges the elite are interested in. This will become even more true over time, as the size of the datasets grows.
A few other items worth considering.
- It’s one thing to construct an internal or private cloud for your own use: it’s quite another to construct an infrastructure sufficient for volume use. And volume use will be a requisite if you’re to make the economics work, because the margins of the cloud business to date have been exceedingly slim. The net? I expect both the barriers to entry and the multi-faceted economies of scale to very actively throttle private cloud construction initiatives. As evidence, I submit…
- The fact that the overwhelming majority of cloud based startups we speak with today have declined to build their own infrastructure, and instead are focused on building value around infrastructures built and maintained by large providers like Amazon, Google, and Rackspace.
CIOs Aren’t Stupid
· They’re bearing the scars of decades of being locked-in by Microsoft on their desktop software, and by Oracle on their databases; the situation where your infrastructure vendors gain control over part of your budget.
So you can bet that as larger outfits take a strategic view of cloud computing, they’re going to be appropriately paranoid about barriers to exit. There’s going to be a powerful demand for standards and for demonstrated interoperability.
Again, I disagree. I’m not arguing that CIOs are stupid, but rather they are constantly under pressure and hugely incented – both organizationally and personally – to focus on the now at the expense of the then. Like baseball owners that get caught up in the moment and invest in massive free agent contracts that later (inevitably) prove to be albatrosses, history has demonstrated to me quite adequately that enterprises, at least, are more than willing to lock themself in if it solves a problem they have now quickly and cost effectively.
To be sure, there will undoubtedly be a great deal of discussion concerning lock-in and the need to avoid it, but I am entirely unconvinced that the industry, generally, has absorbed the lessons of the past sufficiently to avoid repeating it. But maybe I’m just a cynic.
A Historical Analogy
· To me, the cloud-computing landscape feels like the Web search landscape in 1996. Back then, Everybody who’d thought about it had clued in that this was going to be really interesting and useful. There had been a few offerings, much better than nothing but not really hitting a sweet spot. [Disclosure: One of them was me.] Then Altavista launched and it was clearly better than anything else. Meanwhile, Larry and Sergey were at Stanford thinking about transitive functions over the graph of Web hyperlinks.
Amazon Web Services smells like Altavista to me; a huge step in a good direction. But there are some very good Big Ideas waiting out there to launch, probably incubating right now in a garage or grad school.
I’m quite sure this is true; I just don’t believe many of these Big Ideas will involve constructing infrastructures like Amazon’s.
Such an idea doesn’t have to come from a big company. And it doesn’t have to be proprietary to grow fast and dominate the landscape. It has to have no aroma of lock-in, it has to be obviously better, and most of all, more than anything else, it has to be really, really easy to get started with.
Assuming that I’m incorrect, however, and that there will be a vibrant startup market aimed at delivering superior value to Amazon, how likely is it that they survive to challenge the big boys? From where I sit, it seems far more likely that these heavily capitalized startups – hardware, bandwidth and networking cost money, still – will be acquired early and often by the big boys, who are content to outsource innovation in the fabric to startups.
Ultimately, Tim and I agree that the cloud is still waiting on a platform that is truly easy to get started with, much in the way that Google was easier to use than Altavista. I’ll also allow that it’s possible that Amazon goes the way of the one time search giant. I simply don’t consider that likely, and if anything, it seems even less likely to me that we’ll see a cloud provider achieve the kind of market dominance that Google, IBM, Microsoft et al achieved before it.
But it’ll be fun to watch, either way.