Data centers – Part 4 – The future of infrastructure?

I think that proprietary ecosystems like Amazon Web Services will continue to grow.  These ecosystems will be popular with small software developers because it saves a lot of time.  Larger developers will see less benefit from proprietary software because their problems tend to be more specialized and difficult.

The main benefit of pooled infrastructure for larger developers is higher server utilization.  Many of their workloads see fluctuating demand depending on the time of day, depending on the season, or depending on one-time events (e.g. ticketing companies will see a heavy server load when tickets for their most popular events first become available).  They should see their costs drop by taking advantage of the elasticity made possible by pooled infrastructure.

Ideally, software developers will want to have at least two different vendors for their infrastructure needs.  History has shown that having infrastructure at 3 different Amazon data centers does not work because there have been outages that have affected all of the data centers within the same region.  For higher availability, many software companies will want to use at least 2 different vendors in case one vendor has problems.  Open standards and high-quality open source software will make this process much easier.  While it is unclear if open source software will become popular, I think that software developers will find a way to use two different vendors for their elastic computing needs.  This will make data centers more commoditized as large developers will shy away from proprietary data center infrastructure platforms with vendor lock-in.  Small developers may accept proprietary ecosystems because they can save so much time.

To recap:

  1. Serving small customers will be a business where value-added software can generate high margins.
  2. Serving large customers will likely become a commodity business.  However, some of them will require various forms of IT services which can be a higher-margin business.

Amazon is currently the leader in #1.  Who knows if they will maintain their leadership in the future.

As for #2, my guess is that Google may become the lowest-cost operator in a commodity business.  Google has scale, is well-managed, and (less importantly) is currently far ahead of its competitors in finding data center efficiencies.

Centralized management versus traditional colocation

Currently, traditional colocation can be cheaper.  In some cases, all-in costs of traditional colocation are dramatically lower.  In the future, I think that centralized management will offer costs below that of traditional colocation for the most common types of hardware.  It might look something like this:

  1. The rental model is not viable for delivering low-cost infrastructure.  Infrastructure providers will need to look at locking in customers through multi-year contracts.
  2. Data center design will resemble what Google is doing.  Currently, Google’s competitors are lagging behind in terms of PUE and other data center efficiencies.
  3. Server design will resemble what Google is doing.  Currently, it does not seem like many infrastructure providers are buying naked servers with on-board batteries for power backup.  The competition still trails behind Google.  IBM/Softlayer for example uses conventional server designs from Supermicro (SMCI).
  4. Scale.  There are minor cost benefits from negotiating power and spreading out R&D costs over a larger infrastructure base.  The largest players can buy custom hardware or even design their own hardware specifically for data center use.  Google for example designs its own solid state drives and its own in-house Software Defined Networking (SDN) technology.  It also has significant amounts of excess capacity in the “dark fibre” (unused fibre) private network that connects its global data centers.

Currently, nobody does this.  Amazon has the right billing structure in place with Reserved Instances.  However, its prices are too high and Amazon lags behind Google in terms of data center efficiency.  I believe that Google is currently the low-cost operator.  If it decides to price aggressively and attack this market, it could entice many to switch away from traditional colocation.  Google has a lot of excess cash so it’s not unreasonable for Google to deploy it in data center infrastructure and generate some yield.

What will happen to traditional colocation?

I’m not sure because technology is extremely hard to predict.  One trend in favour of the colocation market is the trend towards cloud software that runs on a server rather than a customer’s server or desktop.  As the Internet becomes faster and more prevalent (e.g. smartphones), cloud software will become more compelling.  This helps to create demand for colocation.

Over time, I think that colocation companies will shift towards managing the purchase and maintenance of servers on the customers’ behalf.  Their facilities may become a mix of managed servers and traditional colocation.

  • Scale will become more important as there will be a cost advantage from following in Google’s footsteps.
  • Private fibre networks will be an asset. This goes back to scale being an asset.
  • Smaller colocation providers will likely fare a little worse than the biggest companies.
  • Old data centers will be cost-disadvantaged due to their power inefficiencies.

Business models

In my opinion, the most attractive businesses are the ones that are able to generate very high returns on capital.  Software tends to have the potential for higher returns than IT services, which tends to have higher returns than renting out commodity hardware.  One of the less attractive aspects of pooled infrastructure is that it is very capital-intensive.  As well, there is risk with rental pricing structures from the risk of building too much capacity and not finding customers to pay for it.  This can make the risk-adjusted returns less attractive. While current rental prices are high, price competition may drive margins down and make the business less attractive.

The infrastructure companies that are involved in high-value software and IT services are more interesting.  Those that can convince customers to pay for 3+ years of infrastructure upfront also have the potential of generating higher returns on capital.

Thoughts about particular stocks


Google is a significant user of its own pooled infrastructure.  Other than search, it has a very large number of services running on its own infrastructure (e.g. Gmail, Youtube, etc.) as it currently lists around 42 products.  Out of necessity, it has developed a huge amount of in-house technology as well as buying infrastructure relevant to pooled infrastructure:

  • Google File System
  • Custom SSD drives.
  • Custom switches.
  • Software defined networking (SDN).
  • Private fiber network between its data centers.
  • Data center Cooling technologies.
  • Various patents regarding data center technology.
  • Load balancing software.
  • Google can shift its workload to another data center if the air humidity at a certain location makes its cooling less power-efficient than another data center.

I’m extremely impressed by Google’s in-house technology and expertise.  Selling its infrastructure as a service should give Google a way to sell its in-house technology and expertise to the wider world.

At the end of the day however, I’m not sure if selling infrastructure will move the needle significantly at Google.  I’m far more excited about the potential in Youtube.


Amazon has a head start in pooled infrastructure because it was one of the pioneers.  It has a big lead in software over Google while Google seems to have a big lead in data center efficiency.  Some of its software is very compelling.


Rackspace’s motto is “Fanatical Support®”.  Their vertically-integrated business largely involves selling infrastructure with bundled IT services.

Rackspace is involved in various business lines:

  1. Customers who want a fraction of a server.  e.g. Virtual Private Servers (VPS).
  2. Customers who want a single server.  e.g. managed servers, dedicated servers, and custom servers.
  3. Customers who want multiple servers.  e.g. “cloud” infrastructure.

To me, it seems that Rackspace is constantly thinking about what the future will look like.  They are trying to position themselves to be on the right side of secular shifts in the industry while trying to generate high returns on capital.

In general, they have earned significantly higher returns on capital than the data center REITs that largely rent out space in a building.  Currently, Rackspace’s P/E ratio is rather high (70).  I definitely would not pay such a high multiple for the company.


In my opinion, what SoftLayer does is not that technologically complex or sophisticated as what Amazon is doing with its software or what Google is doing with server and data center design.  In terms of server design, I believe that SoftLayer purchases servers from Super Micro (which makes high-quality conventional servers without brand name HP/Dell pricing).  SoftLayer does not seem to be following in Google’s footsteps and exploring unconventional server designs like Rackspace and Facebook.  SoftLayer is very good at operations but is not on the cutting edge of technology.

I don’t understand why IBM bought SoftLayer.  In the past, IBM sold off its hardware businesses and its semiconductor fabs because those weren’t great businesses.  Now it seems to be doing the opposite and getting back into mediocre capital-intensive businesses.  SoftLayer’s business will likely be fairly commoditized and generate low returns on capital.  I don’t see the potential synergies or the value in their technology.  While purchasing SoftLayer allows IBM to tick off the “cloud” checkbox, it is a poor reason to buy SoftLayer.  I would not be surprised if IBM’s CEO (Virginia Rometty) is soon fired.  IBM hasn’t struck me as being well-managed under her leadership.

In my opinion, IBM should be going after companies that would help IBM provide a turnkey solution for what their clients want to do.  Many of them will likely want to use infrastructure from two different vendors.  If one vendor’s data center goes down, their systems can automatically fail over onto the other vendor’s infrastructure.  IBM should work on providing a turnkey solution (for a very difficult problem) that turns infrastructure into a commodity.  I don’t see why they should get into the business of providing commodity infrastructure.  IBM clients can easily purchase their own commodity infrastructure.  Getting it to work is the challenge and is the reason why they might want to hire IBM.

*Disclosure: No positions.

One thought on “Data centers – Part 4 – The future of infrastructure?

  1. I think one reason why IBM needs an IaaS/PaaS offer (which is capital intensive and has low ROC) is it needs the data centers for its own SaaS. Dropbox can use AWS. IBM can’t for PR reason.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s