July 29, 2013
One of my favorite things to do with kids is to play “word tricks” to make sure that they are listening. The general gist is that you ask a question that seems straightforward on the surface, but is actually nonsensical. For example, “what’s your favorite color: circle, square or triangle?” You’d be amazed the number of kids that answer with a shape. The art of deception is alive and well as I was reminded while reading piece on Data Center Knowledge called: Storage Wars: Dispelling the Myth of Flash Economics
It got me thinking a lot about one of the challenges we have in the “power outlet” side of the business. Not too many years ago (2005 or so), the common metric used for the cost of building datacenters was cost per square foot. Of course, the problems back then when comparing two cost metrics were whether the “square foot” was over the raised floor or the total building footprint. Now, of course, we know that due to varying densities, cost per square foot is an unreliable metric. Cost per kW of IT load is a much better metric, or is it?
You see, slight of hand and plays on words works with this metric too. It is has become all too commonplace for $/kW to omit more than a few meaningful items. What is, and more importantly, what is not included is also very relevant when examining these metrics. Pray tell, Chris, whatever do you mean? When engineers, builders or providers quote their $/kW build metrics, the most common omissions include the building, the land and ancillary supporting space build out (like a loading dock or security station). Some even omit “minor” items like installation and commissioning. I even encountered one ludicrous scenario that provided all-inclusive pricing except for the cooling. That was useful…
Now some of you will be quick to say, “Yeah, that’s true, but the building is a minor cost compared to the datacenter.” In those ancient of days before modular and phased builds (in other words about 5 years ago), you would have a relatively defensible point. When you are building out 10MW of IT load at once, the building cost does become a more minor element (15%-30% of project cost). However, if your end point is 10MW, but you only build 2MW day one, I posit that $/kW is a meaningless metric when the “end point” cost per kW is used instead of the day one cost per kW.
Why would that be? The entire purpose of a modular or phased approach is to manage spending and match load availability with usage. Here’s the multi-million dollar question: If you have the flexibility of growth, as you need it, and that growth doesn’t materialize according to plan, what’s your cost per kW? I have yet to meet a user or a provider that does an accurate exercise on $/kW when presenting to management. You see, everyone presents the end point of 10MW. But the whole reason the C-suite wants a modular/phased approach is to get the option value of not having to spend the money for 10MW. If you only compare end points, it is a false comparison. What happens if that load never materializes? You won’t build, right? So what’s the $/kW then – yup, you guessed it, a boatload more. The unasked question is: What’s the probability of the future load? Don’t you think it makes sense to include that in the analysis?
Let’s use an example to illustrate my point. Let’s say I need to have to 2MW of mission critical load day one. My company is projected to hit 3MW in 2 years, and we think it is prudent to make sure that we can go to 6MW in the long term as we are making a 25-year decision with our new datacenter and that seems like a really good wild-ass guess for the future. So what’s the play? Most experts in the industry would say the following:
• Build out land and shell to support 10MW of utility power and 50k-60k of raised floor
• Fit the shell out with security, loading, some basic office, storage, break areas, and a few telecom rooms
• Build out the first 2 MW of IT load in 17k-20k raised floor with a backplane that can support 6MW
o Leave shell space for the remainder
o Have the ability to add to the mechanical and electrical backplane for 6MW end point in order to make sure it is as “efficient” as it can be at end point
If you came to me with that solution, you would be fired, or at a minimum, severely chastened. Let’s do the math to explain why:
• Land and Shell: You need around 100k sf of shell at $100-$200 psf based upon the level of hardening and cost of utility delivery – let’s split the difference and say $15MM
• Core: Nothing too fancy, but it needs to be good quality. Let’s assume 10k square feet of office, break rooms, loading, corridors and telco rooms. $4MM, but, hey, it’s “only” $40 psf over the whole building footprint…
• 2MW of IT load with a backplane to support 6MW: $14,000 per kW, but it is actually only $7,000 per kW by the time we get to the end point of 6MW. $28MM
So what are the totals day one and at the 6MW end point?
• Total Day 1 price tag: $47MM, day 1 cost per kW: $23,500
• Total end point pricing at 6MW: $61MM, end point cost per kW: $10,167
This little example above, however, is almost always presented as a cost of $7,000 per kW – whether the company is public or private; whether engineer or builder. Why? Is it because the other numbers just don’t look as good? I truly believe that the only way to analyze a build is on what you know at that time. If you are certain that the expansions will occur, then, by all means, do this type of analysis. But if you are honest, you will take into account the day one costs as if you never expand again as well when comparing options. How much is option value worth? Ask your CFO or your customer’s CFO. It would be that it’s worth a lot more than the “efficiency” gained by a pre-configured data center expansion that occurs 3-7 years in the future…
A lot of times, people in our industry complain about the C-Suite not understanding or speaking the language of the data center. I would venture to say that until we start speaking the right metrics of TCO, we are ones answering “Brown” to the C-Suite question of “What’s your favorite shape?”