Highlights from a Cleantech Group panel discussion featuring top innovators, Hydropoint, ChemTreat, Black & Veatch

 “A data center requires about a 500-milliliter bottle of water to generate 10 to 50 medium-length GPT-3 responses,” -- Shaolei Ren et al. University of California

WATERTODAY checked in with Cleantech Group’s Parker Bovée on the alarming consumption of water in data centers –especially from the world’s driest areas and the apparent nonchalance of the consequences.

By Suzanne Forcese

“OpenAI, Facebook, and Google require millions of gallons of water to cool their data centers, municipalities are running extreme risk of devastating water supplies for residents and commercial businesses alike. Data centers currently draw water from 90% of U.S. watersheds, increasing their total withdraws by 35% since 2018. Cities, states, and countries across the globe are committing to this unprecedented water use without clear foresight on what the consequences will be.” -- Parker Bovée

Every Drop Counts

Bovée adds that pathways forward through this uncertainty are required. The realities of resource constraints, public engagement, waste management, and partnerships are the key pillars on which successful data center projects are built.

Power, water, and land availability are core factors guiding decision making. Leaders in data center development follow this thesis closely, most often opting for former power stations or manufacturing facilities.

WT learned more from Jim Oliver, global industrial water supply leader of Black & Veatch:

Elon Musk’s x AI decided to locate the world’s largest data center in Memphis at a former Electrolux factory, partially for its natural gas availability. Similarly, Microsoft is developing two data centers in Leeds, England at former power stations while Amazon announced a data center at a former Virgina power station.

Power stations and former factories are prime targets for data centers for a few reasons. Development timelines can be fast tracked thanks to existing infrastructure such as grid interconnects, transmission lines, substations, bypassing lengthy regulatory permitting processes. Former power sites are also equipped with extensive water infrastructure such as Google’s Jackson County data center, a former coal-fired power plant.

While not a direct conversion, power plants’ location near water resources like lakes, rivers, and aquifers make data center renovation far more convenient than a full buildout.

In Bovée’s analysis public engagement will intervene.

“Communities are highly unlikely to react positively to requests to save water alongside news of data center development. Individual companies will need to focus on reducing waste in water use and find ways to incorporate communities in these efforts,” he says, while adding some examples of current public engagement that are being noted globally.

Announcements for a Greystoke data center on intentionally undeveloped green corridors in Abbots Langley, England alienated entire communities, leading local authorities to reject construction permits. Similar community concerns over power and water use in Santiago, Chile forced delays on Google’s $200M data center investment. Communities are already making massive sacrifices, often forgoing the development of affordable housing to divert water to data centers or semiconductor fabricators. Reoccurring conflict between corporations and communities led to preemptive data center restrictions in Germany, the Netherlands, and Singapore.

Perhaps the best example of successful community engagement in data center development is Meta’s Prineville, Oregon facility. With slow growth over the last decade, Meta was public about expected resource use and encouraged critique from concerned citizens. Over the course of 14 years, the data center has roughly doubled in size with consistent job growth for Pineville through construction and service industries. Issues still exist, especially around water quality and scarcity, but Meta has been especially responsive to community recommendations. Recent critiques on possible water pollution from faulty discharge systems will put Meta’s community commitment to the test in coming months.

There are also some unpleasant difficulties with waste management in data centers that were revealed to WT by Jim Oliver.

Specifically, he identified salt accumulation from cooling evaporation as a key hurdle for efficient operations. Jim was critical of data centers claiming the salt was a valuable byproduct to be valorized, citing salt build up as a critical concern that flies under the radar. Microsoft was forced to redesign part of their Washington data center operations after high brine discharge damaged local water supplies.

Bovée points to some promising developments in reclaimed and recycled water.

Data centers, like any other facility, operate with the limited resources available to them. ChemTreat focuses on this specific area, improving data center water efficiency using pretreatment or recycling of water supplies.

Black & Veatch has helped data centers safely utilize salt or brackish water in their operations.

Chris Spain CEO and Co-Founder of Hydropoint Data Systems addresses a simple golden rule of water operations: stop wasting water. In fact, this is Hydropoint’s key value add. The company focuses on real-time water analytics, identifying water loss as it occurs. Spain is emphatic that existing infrastructure is failing and will need widespread digital support before and after it is replaced, a process that will likely span multiple decades nationally.

Spain made arguments beyond digitalization though. Citing waste as the key opponent, he advocated for creativity amongst players in the water value chain to optimize use. An excellent example of this is Meta’s Gallatin, Tennessee data center. The facility partnered with local waste treatment plants to divert non-potable water to Meta for use. Additionally, Meta equipped their facility with local water recycling and reuse technology stacks, further reducing daily freshwater intake.

Offsetting programs also play a huge role in data centers’ water use. Microsoft partnered with local municipalities, schools, and civil engineering firms to redesign several water systems in the state. While the renovations are designed to offset water use from new data centers, water recovery outpaced data center water intake in many cases, making the collaborations a victory for all parties involved.

Bovée concludes that while it is unclear how data centers will evolve next in their water efficiency upgrades, two major themes to watch are direct-to-chip cooling and new regulatory oversight at both a state and federal level.

Like most technologies in water, regulation is never too far behind the innovation curve so anticipate future conversation emphasizing regulatory control in infrastructure, planning, and water quality issues surrounding data centers.

Related: Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models

The Guardian Revealed: Big tech’s new datacentres will take water from the world’s driest areas