The last quarter of 2011 was very busy for me, so I apologize to you (and to myself) for not being able to hit my personal goal of posting 24 blogs within the one year period from November 2011 to November 2012. It wasn't for lack of trying, but just based on being totally overloaded, with work, conferences, professional and personal obligations and a job search, which, happily, resulted in a new position for me with Digital Realty, the premier and largest data center developer/owner in the known universe. For quite some time, I've wanted to explore the topic of the real value and cost equation relationship in the contemporary data center. The way this blog started was totally different than how it ended, but here goes...

This is a high-level, conceptual exploration of the topic looking at value instead of operational issues or an accounting/audit review of the costs involved and metrics used to evaluate and calculate the relative value and cost of operating a data center.

Certainly, the equation will vary based on the type of data center being considered. Is the facility a "lights out” operation? Is it a Colocation/Managed Services facility"? Is it a Cloud-operation? Is it owned by an "Enterprise”? Each one of these will prompt a different model entailing different costs. All of these facilities and others will have certain features and factors in common, but the operational goal or mission statement of the facility will have implications related to availability, resiliency and the resulting cost.

All data centers use electricity (though that electricity may be created on-site through an unconventional source [think Bloom Box, or micro-turbines]), water (at least to some degree), are located on and in some form of real estate, employ a certain number of professionals (depending on facility type), and have other development and operational costs associated.

Data centers are far from inexpensive to build. Even if one is able to take advantage of existing infrastructure, acquisition and (re)development, capital costs can range from $4,000 to $22,000 per kilowatt, depending on the mechanical and electrical design as well as other structural features. You might be calling "B*[[$+£¥" based on the wide range, so let's accept that the mean cost is about $8,000 per kilowatt, not including acquisition cost. The range fluctuation can definitely be associated with the Tier Level rating created by the Uptime Institute (, or specifically as mentioned above based on mechanical and electrical design redundancy.

Let's assume that to maximize the investment, the facility must contain at least 150,000 square feet and be located on at least 6 acres of land. Let's also assume that we don't want to deal with any existing field or legacy conditions in a building, so we'll be acquiring raw land, or if a building does exist, we'll demolish it. Since we are "server-huggers" we need to be in a major metropolitan area, so our land cost will be higher than if located in a rural area. I'm estimating about $80 per land square foot. So, let's gather this information for our initial pricing model:

Land acquisition cost -                                    $21,000,000

Building shell cost -                                         $11,250,000

Office space cost -                                         $ 900,000

Soft costs (including cost of capital) -            $38,620,360

We usually estimate that about 70% of a facility will be used as "white" space (where the computers are located). Therefore, of the 150ksf, we will have about 105ksf of compute space. In this amount of space, we'll need about 10 - 15 Megawatts of Critical power, that is, power that only supports the IT gear. Development of the white space will likely take place in phases as forecast demand begins to exceed capacity. Based on the above assumptions, the white space or data hall cost of development would be about $120,000,000.

Therefore, the total cost of construction will be approximately $192,000,000. As mentioned above, it’s unlikely that the entire facility would be built initially. Therefore, this is provided as an understanding of the enormity of the overall cost model and for purposes of comparison. Additionally, data center design technology is changing, improving and becoming more efficient, so let’s also assume that we can only amortize this amount over a 10-year period. Based on the current lending environment, it’s pretty safe to assume that the cost of money is about 6%. Finally, we can render a back-of-the-napkin monthly cost of the bricks-and-mortar plus the equipment infrastructure (back-of-house) of about $2,132,000 per month, which is $14.21 per square foot, or $142 per kW.

The recurring monthly costs of operation will include a variety of different line items including power, maintenance, labor and other operating costs. With reference to the phased development and deployment of the data center and IT landscape, the costs associated with the aforementioned line items would also reflect a pro rata incremental adjustments resulting from increased deployment. However, for this discussion, let’s again assume a static load from day one. Therefore, the following costs would apply:

Power (assuming $0.10 per kW) -                                         $1,080,000;

Labor (assuming 20 IT professionals) -                                 $ 166,667 (salary only);

Operating Expenses (including maintenance) -                     $ 300,000;

Approximate Total -                                                                 $1,546,667;

Cost per square foot -                                                             $10.31;

Cost per kW -                                                                                     $103.11.


Some of these costs may or may not be imputed based on the type and operational model of the data center. For example, if the enterprise is operating within a colocation facility, it’s unlikely that there would be a need for all 20 professionals. There may also be some cost savings related to operating expenses due to the scale of the facility and certain economies that can be actualized by the same. Also, design efficiencies will greatly affect the operating costs, including power and water consumption.

If there’s a standard price for colocation per kW that could be benchmarked throughout the US today, it’s probably about $150per kW, gross (net of electricity) and this is for a retail transaction where the power requirement is less than 240 kW. On a wholesale basis, where the requirement exceeds 1 MW, pricing fluctuates, but could be pegged at $120 per kW for quality space, though the market is very competitive and is being constantly commoditized.

Stepping back for a moment to evaluate the above, does it make sense to deploy this kind of capital if data center operations are not core to the business that is operating the facility? The loss of use of capital in and of itself is one primary reason for not building one’s own data center and taking into account all of the other costs involved, it really doesn’t make sense. Factoring in changing technologies that promise to reduce server power consumption and heat production, increased virtualization, and improved cooling technologies will it be long before IT footprints will invert the pricing model and make obsolete the notion of a stand-alone enterprise data center in favor of a hybrid cloud-colocation-managed service model? I don’t have any answers to this question, but based on the new technologies that I’ve seen over the past 6 months, I do know that the data center model is being transformed by a variety of different factors. Stay tuned for my next blog about Facebook’s true and lasting legacy. No, this has nothing to do with how you normally think about Facebook.