The Hidden Costs of AI Data Centers: Power & Water Strain

TL;DR

AI data centers consume enough power to run California for decades and use 5M gallons of water daily, creating massive strain on local communities. #VentureStep #AI #DataCenters

INTRODUCTION

The rapid advancement of artificial intelligence is promising to reshape our world, but this progress comes with a hidden and monumental cost. Behind the seamless interfaces of AI tools like ChatGPT and Gemini lies a sprawling physical infrastructure of data centers that consume astonishing amounts of energy and water, placing an unprecedented strain on our planet's resources and the local communities that host them.

In this episode of Venture Step, host Dalton Anderson dives deep into the environmental and social consequences of the AI boom. He moves beyond the optimistic headlines to uncover the staggering statistics of data center consumption. From power grids pushed to their limits to entire towns' worth of water being used daily, the scale of this resource drain is almost incomprehensible.

Dalton explores real-world case studies, from "Data Center Alley" in Virginia to the controversial build-outs by Meta in Georgia and xAI in Memphis. The conversation highlights the direct impact on residents who face rising utility bills, depleted water sources, and dangerous air pollution—all while the tech giants reap the benefits. This episode serves as a critical examination of the true price of the cloud and asks what we can do to mitigate the damage.

KEY TAKEAWAYS

• By 2030, the energy consumption of data centers is projected to reach 950 terawatt-hours, an amount that could power the state of California for nearly 30 years. ¹¹
• A single mid-to-large-sized data center consumes about 5 million gallons of water per day, equivalent to the water usage of a town with 30,000 to 50,000 people. ²
• Local communities bear the hidden costs of data centers through increased utility rates, as seen in Georgia where some residents' power bills have nearly doubled. ³
• The cooling process is incredibly wasteful, with 80% of the purified drinking water used in data centers evaporating into the atmosphere and never returning to the local ecosystem. ⁴⁴⁴
• To bypass grid limitations, some facilities, like Elon Musk's xAI, have resorted to using unpermitted gas turbines, creating significant air pollution and causing respiratory issues for nearby residents. ⁵⁵⁵

FULL CONVERSATION

Dalton: Welcome to VGSTEP Podcasts, where we discuss entrepreneurship, industry trends, and the occasional book review. ⁶ Today, we're going to be discussing a touchy topic. We're going to talk about AI data centers and the compute load that it puts on both the world and especially its local populace. ⁷

The Shocking Scale of AI's Energy Consumption

Dalton: Currently, data centers in general use up to about 1.5% of the global energy consumption. ⁸By 2030, that energy consumption is predicted to increase to 950 terawatt-hours. ⁹To put that in perspective, one terawatt-hour can power the whole state of California for one and a half weeks. ¹⁰If you do the math, that projected energy usage for 2030 gets you to 28.8 years. ¹¹

So one year of consumption in 2030 is going to be able to power, assuming that California energy consumption doesn't increase, the state of California for almost 30 years, which is mind-boggling. ^12121212

Dalton: The state of California is a big state, big population, lots of homes—the whole state for 30 years, for one year of usage. ¹³I was curious how much energy we are predicted to use in total in 2030, and it was close to 30,000 terawatt-hours. ¹⁴So that puts data center usage at around 3.3%, double what it was previously. ¹⁵I wonder how much higher it's going to go in 2050. I'm not sure, but I do find it very fascinating. ¹⁶I knew it was a lot, but when you write the numbers down, you're like, wait, did I add a zero here? ¹⁷

What Makes an Ideal Data Center Location?

Dalton: In this episode, I'm going to talk about AI data centers, their rapid development, their power usage and grid strain, the pollution contribution, and the construction and water usage costs to local communities. ¹⁸The plan is to build out to 5,000 data centers, and I think there are 3,000 and some change planned. ¹⁹One of the most populous places is called Data Center Alley in North Virginia's area, which has seen massive growth. ²⁰

Dalton: Before this whole AI workflow came about, there were a couple of major cloud providers: Microsoft Azure, Google Cloud Services, and Amazon AWS. ²¹When you're looking for a data center location, you want somewhere that ideally has access to water and good power grid stability. ²²You also want to avoid a place that has a risk for catastrophic events like earthquakes, fires, or hurricanes. ²³You don't want to spend billions of dollars on a build-out near Miami and then a hurricane comes through and floods your data center. ²⁴

Dalton: A lot of times, these places that have data centers built are in areas that have less availability of water or are water-scarce areas. ²⁵The majority of the data center concentration is in areas that don't have that much water available. ²⁶So that's an important note to keep in mind. ²⁷

The Strain on Local Power Grids: A Virginia Case Study

Dalton: I was very surprised that these cities don't have separate policies for the data centers. ²⁸There was just this big rush to get them in there with tax write-offs, benefits, and land. ²⁹ There are some meaningful benefits. Virginia, for example, stated that they're getting $1 billion in taxes from these data centers every year. ³⁰So it produces a lot of tax revenue, but there are also different costs that are socialized among the population. ³¹

One of the costs is energy usage. In 2023, 25% of Virginia's energy usage from its electricity grid is from data centers. And by 2030, it's predicted to be 46% of the total area's energy usage, which I think is quite a sight there. ³²

Dalton: There is a push from local municipalities that they've seen the strain it's been putting on their citizens. ³³They need some kind of guarantee that data centers will make their own energy. ³⁴If you want to build a data center, you need to build out your own power. ³⁵You've seen articles about Google, Microsoft, and Amazon acquiring nuclear startups or licenses to produce nuclear energy. ³⁶

The Hidden Cost Passed on to Residents

Dalton: Another issue is that a third of the data centers in Data Center Alley in Virginia are within 200 feet of a home. ³⁷That is very close. ³⁸I just don't understand the thought process of putting them so close to residents. ³⁹ These are massive facilities. The Colossus data center requires 300 megawatts of power. ⁴⁰To give an example, 350 megawatts would power 350,000 homes. ⁴¹These are massive facilities with large capital outlays. ⁴²

Dalton: Then another thing that's happened recently is these AI workflows from ChatGPT, Gemini, and Anthropic. ⁴³These AI agents or chats use 10 to 30 times the energy of a Google search. ⁴⁴A large data center puts a massive strain on the utilities. ⁴⁵This strain creates a bottleneck for the data centers to expand and it puts a strain on the local populace. ⁴⁶So you have two different groups that are unhappy. ⁴⁷

Dalton: When I was watching a video about a data center in Georgia, it talked about how the biggest power company has had six rate increases and they're not treating data centers differently than they do other corporations. ⁴⁸The cost is being passed on to small business owners and residents. ⁴⁹They're talking about how in peak season, their energy bill has nearly doubled from $250 to $450. ⁵⁰

The Astonishing Water Footprint of the Cloud

Dalton: The next thing is water, and this was something I was very surprised about. ⁵¹

I had no idea that a large to mid-sized data center is using 5 million gallons of water a day. 5 million gallons. That is enough for a town of 30,000 to 50,000 people. ⁵²

Dalton: We're talking about powering 300,000 homes and using enough water per day for up to 50,000 people. ⁵³The scale is just hard to understand. ⁵⁴Mark Zuckerberg was talking about how one of the data centers they have planned for Georgia is going to be roughly the size of Manhattan. ⁵⁵ Manhattan is 13 miles long and three miles wide. That's massive. ⁵⁶

Dalton: Google is at 5.6 billion gallons of water, and Microsoft is at 1.7 billion gallons of water in 2022. ⁵⁷That's a lot of water. ⁵⁸And as I mentioned earlier, the majority of these data centers are in states that are water-strained. ⁵⁹66% of data centers are in areas where water is scarce. ⁶⁰

How Data Centers Waste Water

Dalton: The whole process of cooling the servers is that they're taking purified drinking water. ⁶¹It's not wastewater. ⁶²They're taking drinking water, putting some kind of solution in it to sterilize it so fungus or bacteria doesn't grow, and then they run it through the equipment. ⁶³

The equipment is so hot that it evaporates the water and superheats it. So 80% of the water just turns into vapor and escapes into the atmosphere. ⁶⁴

Dalton: The remaining 20% is dirty, hot water with this chemical solution in it. ⁶⁵When it goes back to the wastewater plant, it puts a lot of strain on the water utility company because they have this massive volume to deal with. ⁶⁶And that 80% of evaporated water goes into the atmosphere and the winds take the water vapor somewhere else. ⁶⁷ It doesn't stay local. So 80% of the water they're using is going to a place that isn't beneficial to the local group. ⁶⁸

Case Study: Mark Zuckerberg's Data Center in Georgia

Dalton: I watched a video about a family who lives 400 yards away from Mark Zuckerberg's data center in Mansfield, Georgia. ^{69696969696969}It was reminiscent of the videos about fracking, where small communities complained about their water lighting on fire and having no water pressure because the fracking companies were using so much of it. ^70707070

Dalton: In that video, they talk about how their water pressure is all messed up and there's sediment in the water. ⁷¹They're using a well, so I'm not totally clear if the data center is also tapping into the well water or if it's strictly pollution from the construction that disrupted the well. ⁷²Another thing noted later on is the cost of their electric bill has doubled, from $250 to $450 in peak season. ⁷³It has put a financial strain on them because they want to retire. ⁷⁴

Case Study: Elon Musk's xAI and Gas Turbine Pollution

Dalton: The situation with xAI is a little different. For Elon Musk's xAI, they're not getting enough power. ⁷⁵To move faster, they had to use these gas turbines. ⁷⁶ These gas turbines emit carcinogens and low ozone pollution that leads to respiratory diseases. The pollution can spread up to 20 miles. ⁷⁷

It's a tough situation where they want to move fast. And it's like 'move fast, break stuff.' But in this scenario, you're moving fast and you're hurting people. ⁷⁸

Dalton: The core issue is that there were no permits. A lawyer requested permits from the city, and the city didn't have anything. 79He went to the Environmental Protection Agency, and they didn't have anything either. 80You can't have these turbines without permits, and you can't have them without any protection against the pollution they emit. 81None of that is being done. 82

Dalton: The stories from the local groups are pretty devastating. To have somewhere you've lived your whole life become a place where you legitimately can't live there anymore is tough. 83The deal was announced when everything was approved and construction had already started. 84People in that area found out about it on X when Elon was talking about how rapidly they're moving. 85It was a real rug pull on folks. 86

Prioritizing Profit, Socializing Costs

Dalton: I saw a comment earlier and I thought it was nice:

These data centers are prioritizing profits and socializing costs. ⁸⁷

Dalton: That was well put. The local population is paying for these things with pollution, noise pollution, light pollution, and disruption to their daily life. 88The AI company is getting all the revenue, but they're not paying their pound of flesh to the local population, either by providing remedies to the pollution they're creating or paying for the power rate increases for everyone in the state. 89The local citizens and activism groups have a voice, but no one's listening. 90They're just screaming into the void. 91

Dalton: In the next episode, we're going to be discussing the benefits of these data centers. ⁹²We'll talk about benefits, increased technology, and different approaches to cooling and being more energy efficient. ⁹³So we'll talk about those things. ⁹⁴But appreciate you listening in today. ⁹⁵

RESOURCES MENTIONED

• More Perfect Union (YouTube Channel)
• Video: "I live 400 yards away from Mark Zuckerberg's data center"
• Video: A report on Elon Musk's xAI data center in Memphis

INDEX OF CONCEPTS

AI Data Centers, Amazon AWS, Anthropic, California, ChatGPT, Colossus data center, Dalton Anderson, Data Center Alley, Doge, Elon Musk, fracking, Gemini, Georgia, Google, Google Cloud Services, Mansfield, Mark Zuckerberg, Memphis, Meta, Miami, Microsoft, Microsoft Azure, More Perfect Union, North Virginia, SpaceX, Tesla, Virginia, xAI