Most data centers were built 7 or more years ago. In 2002 it was only obvious to a few people that the demand for electric power was going to escalate to a point where it became a major factor in the economics of the data center.

It was always a factor of course, but it wasn’t a factor that had people making location decisions, because in those days there were very few scaled out data centers. Yes, there was Yahoo, Amazon, eBay and even Google was starting to “go large.” But in those days, most big data centers weren’t that big and power wasn’t a big part of the cost.

The Cost of Power

In my previous post on this topic, I focused on the issue of location. This time I’ll focus on the cost of power.

Unfortunately, you can’t isolate the cost of power from geographic location. For example, California is a bad place to buy electrical power period. So is New York. In Washington (State), Kentucky and Virginia, power costs are roughly half those of California or New York.

If you add in the international picture then prices vary more widely than that. At one end of the spectrum you have high price Denmark at 22.89 cents/kWh and at the other, cheap and cheerful South Africa at 3.56 cents/kWh, with Canada, Sweden and Finland all coming in at well below the US average of 9.28 cents/kWh. Iceland offers the best deal in Europe and has electricity to spare.

Those prices are averages. From a cost perspective much depends on the industrial discounts offered by the utility company you are dealing with. In the US, industrial users of electricity (and large data centers qualify easily in terms of level of usage) can attract a discount of 40 to 50% above normal commercial levels. That’s huge. For some data centers that level of discount would cut 10% or more from the annual running cost without needing to make any software, hardware or networking component more efficient – or to turn down the air conditioning.

The Longevity of Power Needs

When you’re building a scaled out data center you are building expensive industrial plant. You may profit very well from it, but it will take time to see a return on investment and you’d rather not see the cost parameters suddenly change when you haven’t yet recovered the investment. So the question has to be asked:

Is there any technology development that could suddenly alter these equations?

To answer this question, we need to explain why computers run so hot. For year chips got more powerful by shrinking the size of the circuit that was etched onto silicon and by increasing the cpu clock speed. This worked really well until the semiconductor layer on the chip got so thin that it started to leak voltage in a major way. By 2005/2006 the semiconductor layer was only a couple of atoms thick and it leaked like a seive.

The leaked voltage creates the heat that needs to be cooled. As a rough average about a third of the power requirement of a data center is for cooling the heat that the other two third of the power creates as it dashes through nanocircuits.

Moore’s Law is just about blown now. Clock speeds have reached their limits (because of voltage loss) and miniaturization may only provide another couple of iterations to increase circuit size (i.e. the number of transistors on a chip). So while silicon continues to dominate as the material for cpu chips, the heat problem is not going away.

But what if a silicon substitute was found?

Right now, the smart money is on a form of carbon known as graphine to enable chips to continue to grow smaller, although it’s not yet known if it will be industrializable and if it is, it’s 5 to 10 years away. Nevertheless, graphine would be severely disruptive to all data center cost considerations, because graphine runs cold.

So where does that leave us?

It leaves us with scaled out data centers that would like to recover their costs in 5 years or less.

See also:

The Death of the Data Center: The Model
The Death of the Data Center: Location, Location, Location
The Death of the Data Center: Power
The Death of the Data Center: Cooling
The Death of the Data Center: Networking
The Death of the Data Center: Server Hardware
The Death of the Data Center: The Software
The Death of the Data Center: Software Optimization