What happens when the water landscape sees new faces?

My newsfeed algorithm has figured out that I am interested in data centers, but unfortunately that algorithm hasn’t figured out why yet. It’s possible — likely? — that my interest is peculiarly calculated, like watching a 40 year old US ranching and farming TV drama set in the nineteenth century recently mostly because I could tell that the bird songs heard quietly in the background were correct for the film locations (was that a canyon wren? yes! And then a western meadowlark? even better!). I am sure that some lowly member of the sound crew would be pleased, even if the actors and directors would wonder what kind of nut I must be.

Back to data centers. I will admit that I do wonder if AI is coming for my job, about great power competition over advanced technology, and how our macroeconomic policies need to account for a major restructuring of many sectors, but that’s not what has been keeping coming back. To paraphrase a politician from the same era as the accurate birdsong western TV series, It’s the water, stupid.

I’ve been fascinated to see a kind of profound change happening around water allocation in real time and in just a handful of years, with a completely new category of water user appearing. The implications are important even if you don’t care about AI, data centers, and so-called hyperscalers (the companies who manage and use many data centers). Further, I find the evolution of how we discuss resilience for data centers to be notable for its own burst of change. I’d like to address both topics here, in ways that respect at least some of the complexity at stake.

For instance, the composition of the water community has been relatively stable for many years: we’ve known for decades to look for tension and competition between the energy and food systems. Relevant “energy systems” normally included thermal water-cooled systems like coal, some solar, or nuclear or other water-intensive generation systems like hydropower, while “food systems” have included agriculture, forestry, food processing and delivery systems, and food waste. The food-water-energy “nexus” movement — which hit conversational peak about 2011 or 2012 — was meant to cover the big allocation and governance issues. Indeed, IWRM was supposed to do the same thing before the nexus.

“Data” wasn’t really a category we cared about in water. But that awareness has changed. In that sense, data centers are more comparable to the initial appearance of large water-intensive energy systems in the nineteenth century — coal-fired power became important water consumers in a few countries with the onset of the industrial revolution followed by oil and gas systems and then commercial scale hydropower in the 1870s. Big irrigation, of course, goes back much farther, even millennia — China, Egypt, Mesopotamia, though large-scale commercial irrigated agriculture serving global markets is a more recent innovation. But data centers? They’ve been invisible to most of us.

We treat data centers like a uniform entity, but of course they have important variations, and most of us have a limited ability to classify them. Water is not especially relevant to a number of these variations. They can differ, for instance, by the types and intended purpose of the chips inside that generate most of the heat (and which thus require cooling) as well as the cooling systems themselves (which are normally air or water, with many variations when they do consume water). In some cases, a single data center might use different cooling systems for different operating and ambient conditions. They can be relatively small and compact or gigantic and sprawling, covering many hectares. Their impact becomes especially important when many data centers are concentrated in a relatively small area, such as a city or a single catchment. Roughly speaking, data centers use about four times as much water for energy as they do for cooling. Although an estimate, as an order of magnitude, this provides some context for their water footprint.

Their presence as new water users is especially interesting. An individual data center may not have much of an impact on a particular region or set of other water users. A cluster of them — which is how they often appear — is a different matter. Moreover, they are powerful — and they negotiate with energy and water utilities with political and economic authority. They claim a new seat at the water table, and others must make room. It must have been the same with energy systems in the nineteenth and early twentieth century.

The space they occupy in terms of water consumption, sharing, and governance is not clearly resolved, which is why resilience and data centers is an evolving topic. 

My colleague Josh Weinberg talks about resilience for data centers, resilience in data centers, and regional resilience with data centers. This is a useful framing.

Most of the waterXdata attention looks at cooling water, the relatively smaller amount of water within the data center footprint. And as a result, most of the discussion about resilience and data centers is in these facilities. I’ve already sat through a lot of talks on the subject, primarily about the efficiency of water use, often measured in the amount of heat that can be transferred or the amount of water consumed per chip. These are important talking points and especially useful for the operators of data centers. I am not sure that traditional resilience theory is very helpful or relevant here.

Resilience for data centers, though, is different. How reliable are water supplies for data centers? Do they need to be concerned about competition or scarcity, or perhaps even water quality that may be influenced by flood events? Do they need to ensure redundancy for water security? Can other stakeholders “take” their water? Are multiple cooling systems necessary to encompass a variety of potential states? And of course energy is especially important, which is resilience at a system level. In many countries, we have felt water impacts that have affected the short- and long-term reliability of energy supplies. Indeed, redundancy and flexibility may be as important for energy systems as for direct water supplies. Here, looking at the intersection of extreme weather and extreme operating conditions seems very important. And resilience theory and practice has a lot to say, especially about rare or unusual events (or multiple rare events happening concurrently).

The regional resilience layer is also important from resilience theory and practice: can a single center or cluster of centers have a significant impact on a larger set of stakeholders? Again, mean impacts are probably less important than looking at extreme events, seasonality, and how consumption is prioritized when scarcity exists. Most hyperscalers seem to be working actively with their energy and water utilities, which begs the question of how these groups are collectively seeing risk and resilience. I suspect that they need some support — and potentially some broader feedback too from other stakeholders. Ideally we need to have a shared vision of resilience, which can capture the growth of data centers and their ongoing presence.

I am excited to see the emergence of water and data resilience, actually — even if we have some distance to go, we have a chance to develop a larger, more comprehensive agenda. The water table is growing more crowded, but the resilience table still needs more people to sit down with us.

John Matthews

Corvallis, Oregon, USA