A Social License to Operate

Some folks have questioned my ability to assess the hyperscale data center project in Marana because my expertise is in Mining and not AI. That’s fair. So let me explain my logic.

The data center bonanza currently underway reminds me of a quote from my favorite historical drama on PBS, Elizabeth I. In this drama, King Phillip II of Spain argues with the leader of the Spanish Armada, the Duke of Parma. The Duke is urging the King to hurry up. But Phillip replies:

“Then we must move with feet of lead, lest we stumble.”

Phillip is telling Parma that careful planning is necessary to ensure victory. And then the Armada was wiped out by a storm.

Hyperscale data centers are rushing forward with the promise of victory. They are backed by high-powered billionaire investors who can buy the best lobbyists and lawyers. But rushing too often leads to failure and unexpected calamity, because we didn’t study the consequences; we focused on the sales pitch. 

Hyperscale data centers do not store cat videos or stream Netflix. They are AI centers used for a host of applications like large-language models, surveillance technologies, data mining, price fixing, bitcoin mining, medical technologies, geospatial analysis, and the list goes on. Some of these applications are good, some of them are scary. None of them are regulated. 

The latest computer chips used by the servers installed in these facilities get hot because they draw a lot of electricity. That heat needs to be removed to prevent damage to the chips and servers. Engineers design the server racks with cooling systems that circulate cooling solutions directly to the chips. That solution uses proprietary glycol additives to help remove the heat. In turn, that circulating loop needs to be cooled. This is where the cooling towers, heat exchangers, chillers, and similar equipment come into play. The bigger the loop, the bigger the equipment needed to maintain the temperature of the circulating loop cooling the chips.

Short answer: More electricity = more heat = bigger cooling system = more water.

The rush is real. There are not just one or two hyperscale data centers proposed for Southern Arizona, there were at least 5 at the last count. The total power draw of just the Marana center is enough to power the entire city of Tucson. The total power draw for all 5 if they are built around the same time is currently unknown but some reports indicate these 5 hyperscale facilities may require more than 5,000 MW, enough to power all the homes in Phoenix and Tucson, combined.

As an Engineer, it is my job to carefully assess a project, solve problems, and provide solutions. I have told clients in the past that no problem cannot be solved given enough money, but that does not necessarily mean the investment is a good one. 

A Process Engineer’s most basic skill is balancing inputs and outputs. We call this a material balance. We start by drawing a box. The box represents the process plant. Then we draw arrows into the box, the inputs, and arrows out of the box, the outputs. The number of inputs must match the outputs to achieve a balance. If it’s a metal mine the inputs are water, rocks, chemicals, fuel, air and the outputs are metal, water vapor, gases, and tailings. The tailings contain most of the water and residual chemicals. Additional water is used to reduce dust, drinking water for offices, and fire suppression.

This means that the mine can calculate the amount of water it needs as ‘makeup’ water by balancing the amount of water lost to the tailings.

Data centers are no different. The inputs to the datacenter box are fuel, water, and air, for electrical generation, and chemicals for the chip cooling loop. The outputs are heat, water vapor, combustion gases and spent chemicals from the cooling loop. The amount of water needed to maintain the water balance can be calculated from the amount of water vapor in the output. 

Air cooled systems are being touted as a no-water alternative. But, again, my process box includes electrical generation, so the inputs and outputs are the same. And this time the amount of water I need is larger because the amount of electricity generation needed for an air-cooled system is greater.

The short answer is still the same: More electricity = more heat = bigger air-cooling system = more water.

Water lost to evaporation is a negative on a water balance, like a debit in a checking account, you have to input water from somewhere. When Beale said the data center for Project Blue was water positive I knew they were not telling the truth. You cannot manufacture water, it’s not magic, it’s math. My process box clearly shows this.

Next comes Marana. The initial cost of the data center proposed by Beale was 4.5 billion dollars. But given that the land they purchased is within an active water management zone, they will not be able to take all the ground water they need for make-up water. They would have to purchase CAP water to recharge the aquifer, maintaining the water balance. CAP water costs have gone up 15 percent and may not be available for new users, given our current commitment to reduce CAP water by 760,000 acre-ft each year. That’s a big increase in the operating cost for these data centers. So, they switched to air-cooling. And their capital cost rose from 4.5 to 5.0 billion. Because air cooling is expensive.

Then they explained how much water they would save by turning a cotton field into a data center.

This is a lie of omission. They did not put electricity generation in their data center material balance box. They combined those arrows, water, air and fuel, into electricity. And they never talked about the chemical additive they need for the internal cooling loop. Is the chemical toxic? Does the data center need to be built inside a containment to prevent contamination of the ground water in the event of a fire or other catastrophic event?

The short answer changes to: More electricity = More heat = Bigger chip cooling loop = Bigger external cooling system = more water.

But why the sudden rush?

My background is in Mining. Mining projects are not rushed. They can take decades and require the input of dozens of experts in multi-disciplinary engineering, archeology, environmental sciences, hydrology, chemistry, etc. There are strict regulations to be followed from federal, state and local governments. There are surety bonds required for closure and clean-up so a company can’t just declare bankruptcy and move on, leaving the mess for the taxpayers to clean up. Mining is destructive and a lot of people oppose projects, sometimes for very good reasons. Miners must make their case for why a project is good for a community in public. They sometimes redesign or revise plans to achieve a ‘Social License to Operate’. When they fail to achieve this license, projects are delayed or cancelled.

Hyperscale data centers have none of these. Currently hyperscale data centers have no Code of Federal Regulations, or CFR’s, like a mine does (CFR30 part56), no state or local regulations. In short, there is no protection for the public if a hyperscale data center pollutes ground water, produces heat islands, exceeds noise regulations or is used to spy on us, target us, price fix, or use our personal data to reduce us to commodities instead of human beings. 

And despite what billionaire investors want you to think, they are NOT THE SAME AS HOSPITALS. Such rhetoric is reductive and insulting.  

We can argue the merits of a Mine. But at the end of the day a mine produces a valuable commodity, copper, salt, gravel, gold, silver, and those popular ‘rare earth metals’. Hyperscale data centers have potential and could be transformative but in what way? We don’t know and we should know. We should drive this narrative by regulating AI to ensure it is used to benefit all of us.

For these and many other reasons, I believe we must move with feet of lead, lest we stumble.