Short On Interconnection Or Power Generation?

Previously, in Issue 18 of Interglobix Magazine, I had written the article “Water: Thinking Beyond Our Own Industry” on the fact that adiabatic systems may actually help solve interconnection problems as well as require less generation and save money. At the same time, with the right partnership with the agriculture industry (farmers in particular), we can replenish and neutralize the water usage of the data center and even provide additional water to the community. Since then, I have had many people reach out and ask about the true benefits and economics. Fortunately, I had just done a similar analysis for one of my clients (Western Hydrogen in Arizona) that I could leverage. The following discussion puts real numbers behind the thesis of my previous article—waterless data centers (or data centers that don’t consume water) may not be the right answer for solving the water scarcity problem—especially when power and interconnections are scarce as well.

PUE vs. WUE

First it is important to understand the definitions of PUE and WUE. PUE (Power Usage Effectiveness) is:

PUE =   Total Facility Energy

IT Equipment Energy

This metric is non-dimensional, and the closer this ratio is to “one,” the more efficient the data center. In the late 1990s, PUE was commonly in the range of two to three. Since then, the industry has been approaching “one”.  Table 1 shows the public disclosure of PUE for Microsoft’s data centers by region. Note the variations between them. This variation is due to many reasons, such as climate and whether adiabatic (evaporative cooling) designs are used.

WUE (Water Usage Effectiveness) on the other hand is:

WUE = Annual Water Usage

Total Facility Energy

This figure is typically in terms of liters / kWh and has a minimum value of zero, or no water consumption. Again, these values are shown in Table 1.  As expected, in warm climates (such as Singapore, Arizona, and Texas), data centers with low PUEs have higher WUEs. As an example, when comparing Texas (PUE = 1.28 and WUE = 0.24) to Arizona (PUE = 1.13 and WUE = 1.52), it becomes obvious that even though the climates are similar, the Texas data centers are more heavily relying on non-evaporative cooling while the Arizona data centers are likely using mostly adiabatic or evaporative cooling—and thus substantially more water.

Of course, in cooler climates like Sweden and the Netherlands, temperatures are cool enough to limit the need for water evaporation or non-evaporative cooling, which results in low PUEs and WUEs.

CASE STUDY

For the case study, let’s assume we have a site with 500MW interconnection to support a data center and the decision is whether to build a data center with non-evaporative cooling or evaporative cooling. Using the data in Table 1, we can compare the benefits. The following calculations are based on Microsoft’s data (table 1).

CALCULATING WATER CONSUMPTION PER MW

For sake of argument, let’s assume that the climate conditions are identical in Texas vs. Arizona. Thus we assume that:

• A data center that consumes water and uses evaporative cooling has a PUE of 1.13 and WUE of 1.52
• A data center that uses non-evaporative cooling has a PUE of 1.28 and WUE of 0.24

So, calculating the water usage for the two designs and converting this figure to gallons gives us the following:

• Evaporative: 1.52 liter / kWh x .264 gal / liter x 1000kWh / MWh = 401 gal / MWh
• Non-evaporative: 0.24 liter / kWh x .264 gal / liter x 1000kWh / MWh = 63 gal / MWh

This means that per year:

• Evaporative: 401 gallons / MWh x 24 hr / day x 365 days / year =  3,513,000 gal / MWyr
• Non-evaporative: 63 gallons / MWh x 24 hr / day x 365 days / year =  552,000 gal / MWyr

IMPACT TO IT CAPACITY

With this information alone, and the local pressures on water, one would conclude that the best answer is to select a non-evaporative solution. However, let’s look at the impact to IT capacity and cost of electricity for a 500MW facility.

The PUE in Texas is 1.28 vs. a PUE of 1.13 in Arizona (a difference of 0.15 PUE). This difference equates to 1.28 divided by 1.13 = 1.133, or about 13.3 percent more efficient. 

So, what does efficiency buy you? More critical load for a given interconnection.

In this case, for 500MW, the amount of IT that an evaporative data center can support is:

• Evaporative: 500MW / 1.13 PUE = 442MW critical load
• Non-evaporative: 500MW / 1.28 PUE = 390MW critical load

In other words, you get 52MW of extra IT Capacity for a 500MW interconnection, which in turn delays your need for additional capacity. So, by using water, you get ten percent more IT capacity.

IMPACT TO COST

Efficiency also buys you major savings on electricity costs. At 11 cents / kWh (an average industrial rate in Phoenix), the cost of 500MW per year in Arizona is:

• 0.11 USD / kWh x 1000kWh / MWh x 500MW x 8,760 hr / yr = 482 million USD per year
• 13.3 percent of that is 64,000,000 USD in savings from using evaporative cooling.

However, we still need to quantify the cost of utility water. Table 2 shows the typical water rates in Phoenix, Arizona. For sake of argument, we will use 5.65 USD + 0.62 USD to have an average cost per unit. This amount equates to 6.27 USD per 748 gallons, or 0.00838 USD / gal. So, calculating the water cost difference between evaporative and non-evaporative is as follows:

• Evaporative: 3,513,000 gal / MWyr x 0.00838 USD / gal x 500MW =  14.7 million USD
• Non-evaporative: 552,000 gal / MWyr x 0.00838 USD / gal x 500MW =  2.3 million USD

To calculate the cost of using additional water for evaporation in order to get the actual cost savings, we do the following:

64 million USD – (14.7 million USD – 2.3 million USD) = 51.4 million USD in savings by using water

WATER REPLENISHMENT

Of course, there is the problem of water scarcity and community pressures on rates for both electricity and water. As a result, some of the hyperscalers are building data centers that are less efficient but don’t use water.  However, as outlined in my aforementioned Issue 18 article, there are ways to counter the scarcity—by collaborating with the agriculture industry and helping farmers move from flood irrigation to drip irrigation which can save 50–85 percent of the water needed while also increasing yields. Companies like NDrip provide drip irrigation to farmers that are funded by the replenishment credits data center operators use to offset their water consumption (thus neutralizing their impact).

Of course, this process is an additional cost that erodes the cost benefit of using evaporation, so it is worth calculating as well. Assuming a cost of 0.004 USD / gal (which is on the high side), the calculation for the cost of replenishment credits for a 500MW evaporative data center would be as follows:

0.004 USD / gal x 500MW x (3,513,000 gal / MWyr – 552,000 gal / MWyr) = 5.9 million USD

So, to neutralize the water consumption of a 500MW data center costs 5.9 million USD. However, some hyperscalers are overbuying replenishment credits for good will with the community by matching with another 50 percent. For this case, that would raise the water replenishment cost to 8.9 million USD. Thus, the actual savings for an 500MW evaporative system is:

51.4 million USD – 8.9 million USD = 42.5 million USD in savings per year by using water

As a side note, the Western Hydrogen project mentioned earlier has a 28 million acre-foot aquifer, of which 8,320 acre-feet (2.7 billion gallons) is guaranteed each year for one hundred years—which is well over the needs of a 500MW data center. In that case, since it has its own water, the cost is negligible (drilling wells and running pumps), so the saving there would be 64 million USD since water is in abundance in that part of Arizona.

CONCLUSION

By using water for evaporative data centers for low PUE, the case study of a 500MW data center has demonstrated significant benefits when paired replenishment credits sourced through drip irrigation.

These benefits are:

1 52MW of extra critical IT load. At a time of power constraints, this additional capacity is more valuable then ever. Why would you not strive for the lowest possible PUE?

2 42.5 million USD in energy savings each year, including the cost of 150 percent water replenishment credits so that the community gets an abundance of water as well.

3 51.4 million USD in energy savings each year for a site that has its own water, like Western Hydrogen.

These benefits should really be considered for future data center builds, and I hope it drives data center developers to buck the trend of going waterless when there are other ways to make it a win-win for all stakeholders. Finally, this article is not meant to be comprehensive but rather should provide some insights to drive technical curiosity for people to do their own calculations.   Some obvious omissions include things like the CapEx savings for evaporative systems, water treatment costs and other nuances that were not covered here.  There are also many other solutions, and I hope this article gets others to share. Also, there was no attempt to look at the embedded water in electricity since power plants use significantly more water than data centers, which would make the case for evaporative designs even stronger. At the end of the day, I just want all of us to “think bigger!”