I have already written on how Zynga (NASDAQ:ZNGA) and other large consumers of hosting capacity can use Amazon.com's (NASDAQ:AMZN) AWS in a way that subjects Amazon to adverse selection. I will try, in this article, to explain a bit better how this phenomenon works, and how, if Amazon allows it to go on for long, it might destroy AWS' profitability.
Basically, in the past Zynga used AWS for its hosting needs, but at some point decided to create its own data centers, relying on AWS only for peaking needs. Below I will show the broad implications of this for Zynga and AWS.
Zynga Using AWS For All Its Hosting Needs
Let's say that Zynga had a theoretical load profile that looked something like the chart below (all the numbers are purely for illustrative purposes):
What we see here, is that to handle this activity AWS would need capacity able to service up to 120 transactions in a given time period. This capacity would see a usage factor, over the entire observation, of 46%.
Zynga Decides Handling the Stable Load Indoors, Using AWS for Peaking Needs
So what happens when Zynga decides to bring most of its stable load indoors, into its own data centers, while relying on AWS just for peaking needs?
We can simulate the outcome. Let us say that Zynga decides to set up the capacity to service up to 60 transactions in a given period, or half what was needed before to be able to handle any scenario. This is what happens:
Suddenly, Zynga has its new capacity running at a 77.8% usage factor - a very high usage factor, so the investment is optimized. And AWS, having to provide as much capacity as Zynga but handling just the peaks, will see a usage factor of just 14.2% on that capacity - its own investment will be far from optimized.
Now, AWS will partition capacity over many different customers, some of which will have peak needs at different times, so this mitigates the problem. But remember, AWS will be arbitraged as an outsourcer for the entire capacity, so it can't price for peaking needs, yet will be servicing these peaking needs. What these peaking needs mean is the need to have a large investment in capacity that might sit idle for long periods, so that it's there when needed.
In the power industry, these economics are tackled by power prices being much higher when there are peaks in demand, to compensate the peaking units for the fact that they'll just run for short periods, and sit idle most of the time. Until the hosting industry moves to such a pricing scheme, it's likely that it will fall victim to adverse selection where its customers will take advantage of the phenomenon I described.
Two conclusions to be drawn from the phenomenon I described:
- First, other large AWS customers will be strongly incentivized to do like Zynga once they achieve enough scale and build out their own data centers to handle most stable loads while offloading peak loads to AWS. This will give them better economics, because AWS can't price its services for the high usage factors they will achieve. This will also make for recurrent newsflow where large AWS customers set up their own capacity and leave AWS;
- At the same time, AWS will be turned into a peaking plant having huge investment needs to service peaks in demand, while having to price its services for stable load to attract the customers in the first place. If in one hand AWS can mitigate the problem by having a diverse customer base, on the other this customer base will be geographically close and will tend to have peaks in demand at about the same times, thus negating some of the benefits of diversification.
These conclusions seem to point toward AWS being an unattractive business over time, always subject to adverse selection and high capex needs. As an aside, Rackspace Hosting (NYSE:RAX) will probably suffer from the same phenomenon, and indeed looking at its latest 10-Q bears this out, with puny free cash flow - especially if we ignore the large contribution to operating cash flow coming from accounts payable.
Disclosure: I am short AMZN.