There isn't enough water for all of us
On the importance of fairly negotiating data center contracts
I’ve been writing about AI data center-driven water scarcity for some time. The most common response to my work is - “water is never lost to the earth.” Which is True. But we don’t live on the earth. We live in communities. So while the earth may have enough water to support all of humanity, your community may not have enough water to support the people living in it. Just as we grow enough food on the planet but whole societies are suffering famine, the existence of water on the earth does not protect your community from water scarcity. Water scarcity is driving migration into cities, further exacerbating the pressure on municipal water infrastructure.
So, how do we protect our water? Or at least, slow down AI companies’ gluttony until they build technologies that can sustainably address their water needs at scale? To start, we need to put municipal leaders in a position to fully negotiate data center contracts.
All contracts are negotiable
I had a conversation with a friend recently about a horribly onerous contract she signed for a speaker agent. The contract either required up front payment of all fees or it charged a higher fee plus interest for monthly payments. This wasn’t a lease agreement, credit card, or mortgage, it was for someone to help get her talks. And, of course, after my friend paid in full, they got shoddy service (ignored, talked over, genAI generated speaking applications, etc.). One of the first things I learned in running my own real estate/civil litigation firm was - never sign the other party’s contract. It always sucks.
Intentionally so. We’re taught in law school to create contracts that are completely one-sided towards our client, with the understanding that opposing counsel will do the same. The idea is that you both start at opposite ends and negotiate your way to some sort of middle ground. Which is why you will always lose if you sign someone else’s contract. You have to bring a counterbalance to the table so you can start negotiating your way to a middle ground.
So what does all of this have to do with data center water consumption? Well, everything. Water is a municipal concern. It’s managed at the local level by cities, counties, and special water districts. But I have yet to meet a local official who had any understanding of how much water the data center they approved would require to operate, or how much it would ultimately consume.
We’ve been signing tech company contracts, with no counterbalance. And they suck.
So how did we get into this situation, and how do we remedy it? Math. In order to negotiate, our municipal leaders need to have some understanding of how much water the data center will require before approving the build - during the permitting and zoning process. Which is hard. Well, it was hard.
State of water loss calculations
If you search for how to calculate cooling tower water loss, you’ll get a host of articles which all reference the Makeup Water calculation:
Makeup water = Drift loss + Evaporation loss + Blowdown lossMakeup water is the water added back to the system to compensate for losses. Think of it like tire pressure. As you ride your bike, or drive your car, air leaves the tire for a variety of reasons (heat, pressure, holes, etc.). If the tire loses enough air it can suffer catastrophic failure, so you want to refill the tire to its optimum level regularly. That’s the principal behind makeup water. The cooling tower, and all of the systems that rely on it, risk suffering catastrophic failure if the water level gets too low. So you want to refill the water to its optimum level regularly.
The makeup water calculation considers all the ways cooling towers lose water. When you add all of those losses up, you know how much water you have to put back into the system to keep it at the optimum level. While the base calculation doesn’t look too daunting, the complexity is in the loss calculations.
Drift loss is the water that the wind blows away. It’s heavily impacted by the shape of the cooling tower (some let in more wind than others). But wind is too variable and unpredictable for a granular calculation. So drift loss is usually calculated as a percentage of the water that’s being circulated through the cooling tower.
The complexity ramps up when calculating evaporation and blowdown loss.
Evaporation loss
Evaporation loss is the water that is evaporated during the heat transfer process (see The Fallacy of Closed Loop Cooling Systems for more details). Here’s the evaporation loss calculation:
This calculation requires you to know:
The rate of water circulation (C)
The difference in water temperature from the bottom of the tower to the top of the tower (Ti-To)
And how much heat was added to the water during the heat transfer process (Cp)
None of which is available when a data center is being planned.
Blowdown loss
The blowdown loss calculation isn’t particularly complicated, but it also relies on specific systems operation data. Blowdown loss is the water that is removed from the system because it’s become to concentrated with sediments. Think of it as the dregs - what’s left at the bottom of a coffee pot or a bottle of (unswirled) red wine. The sediments are super concentrated in that last glass. The same happens in engines. It’s why regular oil changes are a key part of car maintenance. As the oil cycles through the engine some evaporates due to heat, and it picks up contaminates from the engine. Over time, the oil gets thicker and starts to concentrate into a sludge. Sludge doesn’t lubricate the engine well, so your car starts to perform poorly.
Blowdown is water that is heavily concentrated with solids that were picked up as it cycled through the cooling system. Data center cooling systems are more sensitive than car engines so, instead of waiting for the water to form a sludge and changing it all at once, data center operators expel some amount of blowdown regularly so the system is never impacted.
Your company comes with a guide for when you should get oil changes based on the amount of miles you drive. Those miles are a proxy for the amount of times the oil has cycled through the engine. The blowdown calculation is based on the same principle. The core calculation is the cycles of concentration (CoC):
Blowdown = [Evaporative loss – (COC – 1) x Drift loss] /(COC – 1) To calculate blowdown, in addition to evaporative loss and drift loss, you need to know the cycle of concentration for the cooling tower. Which is where things get trick. There is no proxy for how many times the water cycles through the system. Instead, the cycle of concentration is calculated one of two ways
By the ratio of chloride content in the circulation water v makeup water
By the ratio of conductivity of the system water v makeup water
Neither ratio is derivable when a data center is being planned.
While the makeup calculation looks straightforward, it is technical and confusing in practice. It relies on specific systems operation data. Data that isn’t available during the data center planning process and is closely held as proprietary information once the data center is in operation. Data center operators are the only people who have access to the data required for these calculations and they currently have no legal obligation to report any of this information at any time.
So what do we do when we’re outside looking in? How do we obtain the information necessary to level the playing field in negotiations? As I mentioned in this IEEE/OECD session during the France AI Summit - we move forward by recognizing the audience and the purpose of the calculations. Municipal leaders don’t need scientifically accurate conclusions, they need directional forecasts. Thankfully, there’s math for that!
Energy-based calculation
The one piece of data we always have about a data center at the earliest planning phases is the electricity it will require. Which is precisely what Uptime Institute utilizes as the key input when for its cooling tower makeup calculation. Uptime Institute created the Topology Tier Standard as “a performance benchmarking system to help data center owners and operators identify the performance capability of the data center infrastructure.” Data centers are ranked in tiers based on their performance on the Standard’s benchmarks. Data centers can be certified at the design level, or while in operation. The makeup calculation for design-level certification utilizes the following assumptions:
Each 1,000 kilowatts of cooling load requires 1,027 US gallons per minute of condenser water flow through the evaporative cooling towers
Cooling system utilizes 3 imperial gallons per minute of condenser water per ton of cooling
Evaporation consumes about 1% of condenser water flow
Drift and blowdown consume about .5% of condenser water flow
By tying water flow to cooling load, Uptime Institute made it possible to calculate makeup water based solely on the amount of energy used for cooling:
The only variable in this calculation is cooling load. The current assessment of the amount of a data center’s energy that is used for cooling is 40%. With all of the assumptions accounted for, we can do the math!
Let’s use the Stargate AI data center campus in Abilene, Texas as an example. The data center is intended to run on 360 megawatts of power. How much water would a data center that runs on that amount of power use and consume?
360MW x 40% = 144MW cooling power
divide by 1000 KW = 144KW
144KW x 1027gpm = 147K gallons used per minute
147k gallons x 60 = 8.8M gallons used per hour
8.8M gallons x 24 = 212M gallons used per day
212M gallons x 1.5% = 3.2M gallons makeup water per day
3.2M gallons x 30 = 95M gallons makeup water per month
95M x 12 = 1.1B gallons makeup water per yearBased on these calculations we can estimate that the Abilene, Texas Stargate AI data center campus will require 212 million US gallons per day to cool. Of that 212 million US gallons, 3.2 million gallons will be consumed. Consumed means lost to the community it was taken from. The evaporated and drift water will move through the global water cycle and the blowdown will need to be treated. There are systems to minimize drift and reuse some percentage of blowdown, but they are expensive. You can’t assume companies will implement them if they aren’t required to. But that’s the point of this exercise, to empower leaders to start those mitigation discussions as early in the process as possible.
DIY Water Loss Calculator
Now that we have a publicly accessible calculation - let’s make it public!! AllAI Consulting, Inc. utilized the Uptime Institute’s Tier Standard: Topology makeup water calculation to create the world’s first public Data Center Water Consumption Calculator (available here).
The estimates provided here can serve as the starting point for discussions on data center water use and consumption. Understand that the amount of water that is requested from your aquifer is not the sum total of the water required to cool the data center. Most of the water will not come from your city’s aquifer, but it’s coming from somewhere. And given the cost and complexity of piping/shipping water from other states and regions, it likely isn’t coming from too far away. Having a sense of the overall system requirements places you in a position to assess the validity of an operator’s water resourcing plans. Having an initial estimate enables you to ask more consequential questions early in the process. Specifically:
💡 How do the assumptions in your estimate differ from those in the tool?
💡 What is your average facility-level water consumption percentage?
💡 On average, how many hours in a year do your facilities run at peak usage?
💡 What is your local water supply chain and what percentage of the estimated usage can it meet today?
💡 How much of the estimated consumption do your replenishment efforts account for, and in what timeframe?
While all contracts are negotiable, drinkable water is finite. There isn’t enough water for all of us and all of the AI data centers companies want to build. Not given the current technology. But there doesn’t have to be. Municipal leaders are charged with using zoning and permitting processes to ensure there’s enough water for the population first. Now that this tool provides communities a legitimate seat at the negotiating table, it’s critical that we secure our survival first. It is a company’s job to secure the resources they need to be successful, not humanity’s. If they cannot operate their businesses within the reality of the drinkable water that exists on the earth - then perhaps they should innovate. Faster.