Water: Thinking Beyond Our Own Industry

As data center growth increases to meet the needs of AI, the scale is significantly stressing resources. Energy is the most talked about topic, with data center operators scrambling for interconnection to the grid for their ever-increasing loads. Similarly, long lead times for equipment have not made the job easier. Even labor is becoming a constraint in some locations.

In addition to all of these challenges, and perhaps more importantly, the volume of water needed for cooling the new and larger data centers is increasing with their scale. Of course, this is a problem in water-stressed, or soon-to-be-stressed, communities (such as Phoenix, Arizona) where many of the hyperscale data centers use adiabatic cooling (think of swamp coolers) to knock down temperatures for GPU and CPU servers. But this dilemma isn’t remotely limited to these areas. Indeed, if you are thinking “data center,” you also need to be thinking “water.”

Of course, engineers like a great challenge and are looking for ways to mitigate the need for water by going to mechanical cooling like chillers to eliminate the need for evaporation. While I applaud the effort, I don’t like the result. By even their own admission in press releases, the PUE will be higher without evaporation.

This quandary illustrates the bigger problem at hand: How can we trade off the tension between better energy efficiency and water consumption? The answer to that question is not easy, but, if we as an industry are prepared to think much more broadly, I believe there is a solution that can do both—achieving optimal PUE and best water outcomes.

The Value of Thinking Broader and Bigger

Since my retirement, I have had the privilege of advising dozens of companies, nearly all related to the data center supply chain. However, my curiosity and obsession with large systems and ecosystems had me explore what seems to be completely out of my wheelhouse: the agriculture industry.  You may ask: What is a data center engineer doing working in the food and fiber production space? How is this a good use of my time? How can I even add value? Well, the answer proved to be deceptively simple: Think bigger and bring diversity of thought. 

With PUE’s success over 20 years ago, we leveraged the concept at the Green Grid and developed a metric called WUE (Water Usage Effectiveness). While this metric did not catch on for about a decade, it was really the beginning of the recognition that water will be a constrained resource we need to care about. However, my real wakeup call was around 2018. I had to delay opening a data center in South Africa because the community we were building in had 25 days of drinking water left. This starkness sensitized me even further as I saw this situation as a “canary in the coal mine.” I saw it not as an annoyance (even if I was annoyed), but as a gift. It provided a view into the future of what might be. Could water be something that limits growth and provide negative perception of our industry? Will we lose goodwill in the communities in which we operate? Can we actually build models where goodwill grows and water becomes plentiful in the future? Let’s face it: Without water there is no future.

So, needless to say, water became top of mind for the R&D projects. Now, seven years later, with water still on my mind, I came across an irrigation company called N-Drip (www.ndrip.com). As its name states, it is a drip irrigation company, but an unusual one, and for at least two reasons.

First, the company is mission-driven like a not-for-profit, focused on solving the water scarcity problem the world is facing. How? By replacing flood irrigation with gravity-powered drip irrigation. With 85 percent of global agriculture (and 40 percent of US agriculture) using flood irrigation, a transition away from flood irrigation to drip irrigation could save about 50 percent of the water consumed in irrigated agriculture while increasing crop yields by 30 percent. Both of those, at scale, offer enormous, transformational change in agriculture.

Second, while solving one problem (water wastefulness), N-Drip doesn’t do so at the cost of something else, like producing carbon or taxing the grid. As N-Drip operates only with sustainable, always-on gravity, the benefit doesn’t create some new problem. But, you may ask, if it is so great, why isn’t every farmer rushing to use it?

The answer is that farming is a very low margin business. Water is mostly free or virtually free. And when water is free, there is little incentive for farmers to help save the world if it comes at the cost of their economic well-being. This dilemma is where the interplay of data centers and agriculture gets interesting. 

Coming from Microsoft, I knew that our goal (and that of nearly all other hyperscalers) was net zero water. Talking with N-Drip, I saw a model where hyperscalers could work with a company like N-Drip to pay for water replenishment credits that cost far below the alternatives and, through N-Drip, fund farmers to install gravity-powered drip irrigation. In return, the many millions of gallons of water saved with the switch from flood irrigation is credited to the hyperscaler or data center operator. N-Drip finds the farmers and installs the system, and since agriculture is a huge industry, the replenishment credits can be achieved at scale. In some cases, the operators even over-buy relative to their usage and return that water to the community (i.e., 150 percent of their actual usage) at a cost that is a multi-decimal point rounding error of the energy cost. It is a win-win-win situation: The data center benefits, the farmer benefits, and the watershed from which the water is drawn benefits. And, to be clear, the data center becomes a local hero—or certainly not a local villain. So, how does this potential tie into our earlier discussion on the tradeoff of efficiency versus water consumption? We realized that eliminating water evaporation for cooling in data centers results in an increase in PUE while decreasing the IT load for a given interconnection. For example, for a 100MW interconnection, a PUE of 1.25 equates to an IT load of 80MW for a waterless data center. By contrast, if you used water evaporation for cooling, the PUE could be as low as 1.15, which equates to 87MW of IT load. This situation means that by using water as the vehicle for cooling, we can get nine percent more IT equipment for a given interconnection.

The importance of this increase is clear. With the scarcity of interconnection to the grid, it must be our goal to get as much load per interconnection as possible—and adiabatic cooling with water evaporation enables that. Here is the beauty: Using drip irrigation to offset the water needed can actually help our industry get more with less (more IT for the same interconnection) and help the agriculture industry get more with less (more crop with less water). Plus, this approach offers more water for nearby towns and cities, and more water for the environment.

A Real Opportunity or a Shell Game?

Some people might argue that water replenishment adds cost or that this approach is a mere shell game of moving water from one industry to another. So, let’s address both of these concerns.

Cost is always a big deal when operating at the scale at which we are operating. Ultimately each company will have to do its own TCO and due diligence, but here are some basic points to think about:

1. Depending on PUE, the penalty for a waterless data center would likely be in the range of 3–9 percent more energy cost for the same load.

2. Your data centers would need more mechanical cooling, which adds cost.

3. You would need to find 3–9 percent more interconnection capacity, which likely means having to build more data centers.

As far as the argument that this approach is a shell game, people could argue that the data center would essentially be consuming water that would have been used by the farmer. However, since the cost of the water is orders of magnitude lower than the cost of energy, most of the end users are effectively over-buying replenishment at 150–200 percent. So, this approach effectively meets the data centers’ needs while simultaneously giving back to the community and to its watershed.

Additionally, space doesn’t permit a full discussion, but depending on the local power utility mix of energy sources, it could be that the water used to produce the electricity for the air cooling is actually a larger amount at the power plant (because of higher PUE) than the amount of water that would be used at a data center using evaporative cooling. If so, that would really be a shell game!

Final Thoughts

So, the key takeaways from this article are:

1. There are huge opportunities if we look beyond our normal industry boundaries. The case argued here is that by understanding what is important to us and what is important to the agriculture industry, we can find win-win situations to provide a more sustainable future.

2.  We should be optimizing more broadly for the larger system. As in the case of “waterless” data centers, sometimes our intuition isn’t always the right one. In this case, when the industry is short on interconnection capacity, we should be optimizing for lower PUE and finding ways to make up the water.

Finally, my goal of this paper is to stretch our thinking. I would encourage readers to join me in this analysis and to broaden the conversation. One way or another, our industry needs to be thinking a lot more about water—and thinking bigger and more broadly! In this case, use water, save more water and get more IT load for a given interconnection to the grid.