The Environmental Impact
Part II will cover the financing of data centers and Part III will cover the economic aspects of data centers specifically and technology generally.
Actual Bipartisan Consensus
Data centers are much in the news these days. I should say more precisely, opposition to data centers is in the news today. This is a rare case of bipartisan consensus in that both parties have major figures taking stands against the construction of more data centers. There are, in a growing number of states, laws and regulations being put into place to impose outright bans, moratoriums, or stringent regulations governing the construction of these centers. Along with these restrictions many states have cut back on tax incentives, grants and other subsidies that serve to encourage data center construction. Additionally, many states are requiring data centers that are built to provide their own power source. A listing of these and where they are in the legislative process can be found here. This list just covers the legislative process. It does not include those states whose governors have taken executive action. Texas—yes Texas—has put significant obstacles to new data center construction via an order from Governor Greg Abbott. This entails an audit of all data centers in the ERCOT service area going through ERCOT’s interconnection process. This audit requires a disclosure of all tax breaks a project will receive. This is a topic further explored next week in Part II. Even my deep red home state of Nebraska has pulled back incentives and made it generally more difficult to build data centers. On top of all that, many municipalities and counties have put in place data center restrictions.
This follows from a cratering of public support for data centers. According to Gallup, 7 in 10 Americans do not want data centers built in their localities. The top reasons are concerns about the effect on natural resources, especially water, quality of life concerns (pollution, including noise), and effects on costs of energy due to increased energy demands. The polls are clear, but is the public? Are the stated concerns valid or is this a misguided moral panic?
What Are We Discussing?
AI data centers are essentially akin to heavy industrial facilities, at least in terms of their physical structures and footprints. The most comparable types of facilities would be microchip fabrication plants, smelters and steel mills. In many ways data centers have a smaller impact than these types of uses. This is true certainly in the case of land and water usage and air pollution. It is not that these aren’t legitimate concerns. It is just that the mainstream press has tended to exaggerate these concerns. Also, most stories about AI data centers have only discussed some potential drawbacks. They have failed to also consider the upside benefits of what AI can offer the economy.
The real problem with AI data centers is not their physical and environmental impact, but the way they are financed, which is the subject of next week’s essay. If one gets bogged down in environmental objections, which are primarily just NIMBYism, then one would miss the real market distorting dimension of the current financing regime. This post covers the environmental effects.
The Physical Footprint
This is a detailed topic that can glaze one’s eyes over very quickly. I think it will be easiest to offer a brief comparison of the physical and environmental impacts of data centers compared to both office buildings and heavy industrial uses. As will be obvious, the closer comparison is between data centers and heavy industrial uses.
As far as water usage is concerned data centers use much more water than office buildings but about the same as heavy industrial uses. The land use of a data center is higher than that of office buildings but roughly the same as heavy industrial uses. Emissions from data centers are a bit higher than office buildings but much less than heavy industrial uses. Electricity usage is much higher in data centers than office buildings but about the same as heavy industrial facilities. The employment impact is much less in data centers than in either office buildings or heavy industrial plants.
The water usage information can be found here, and here (note this USGS study is dated, but the most recent available). Power consumption information can be found here.
It is necessary to place all of this in context. It is therefore helpful to compare AI data center usage to other uses of environmental resources such as water and electricity.
When discussing water usage there are two concepts to keep in mind: withdrawal and consumption. Withdrawal is the amount of water that is removed from a particular source. Consumption is the amount of water that is used up and not returned to the environment. Therefore, the withdrawal amount will always be larger than the consumption amount. How much water is returned to the environment depends upon the use. In order to make an evenhanded comparison I will use the amount of water consumed, that is the amount not returned to the environment. This seems reasonable, as the concern is about the effects of data centers upon the environment. Also, to be even more than fair, using consumption figures will present the maximum water consumption for data center use. Additionally, to be maximally fair, this will include the consumption of water by energy producers who power data centers.
The largest consumption of water is by the agricultural sector, including livestock and aquaculture. They represent 82-88% of U.S. water consumption. This is because they consume 60-70% of what they withdraw, much in the form of irrigation.
Industrial uses, which includes thermoelectric power generation make up about 6-10% of U.S. water consumption. These uses return 75+% of withdrawals back to the environment.
AI data centers currently use about 0.6% of total U.S. water consumption. Of that, 0.05% is about 17.4 billion gallons of direct consumption onsite. Add to that about 211 billion gallons consumed by power plants generating data center electricity, twelve times the onsite number. Note that data centers indirect share of water consumption is drawn from the industrial figures, not added to it. Also, this is power plant water, not strictly data center water. This will certainly grow while remaining a much smaller percentage than most other uses of U.S. water resources.
Data centers also consume most of what they withdraw, about 80%. Fortunately, new “closed-loop” systems are being deployed that consume only 5-10% of the water withdrawn to cool the centers. Air-cooled facilities essentially consume no water.
Golf courses and landscaping account for about 490 billion gallons of water consumed per year, roughly 1.3% of U.S. water consumption. This means that currently this usage is about twice the usage of data centers. Additionally, data center usage will not eclipse this anytime soon.
As for power consumption, data centers use just 4.7% of U.S. electricity consumption (although geographically concentrated). This is projected to rise to possibly 12% by 2030. Even if that is the case it is still a smaller share of power consumption than the electrification of transport and heating will increase over the coming decades.
Surprisingly this has on net lowered rates. In fact, the rates would have been about 6% higher without the data centers built between 2019-2024. Briefly this occurs because retail rates must recover average costs, not marginal ones. So, a large steady new user of power spreads these large, fixed costs across more kilowatt-hours generated, lowering average costs. This steady demand pulls forward new generation which will be less expensive than what it replaced.
The concern is that new sustained demand could lead to price increases as new generating capacity is built, but this price signal to industry to increase the capacity of the grid will strengthen a grid that has become shaky under the demands of greater electrification. In time this build out will lower average prices as it has been doing up to now. That is not to say that there has been no shifting of costs of energy usage to other ratepayers, particularly in public utilities. This will be discussed next week in the context of financial subsidies to the data center industry.
One additional impact is that of noise pollution. AI data centers are loud inside next to the servers, comparable to a heavy industrial facility. Most people in nearby neighborhoods would hear something in the range of 40-59 db. This is a constant presence but not hearing-damaging. Proper siting of these centers in relatively remote areas can mitigate these noise effects without hindering the construction of data centers.
The other noise related objection surrounds the fact that generators undergo load-tests that are loud and continuous for a significant time. However, this testing is scheduled. It can be mitigated by scheduling at times less likely to be intrusive. The other noise and air pollution related issue is the use of diesel powered backup generators. They can be loud but proper mitigation techniques can reduce this noise and air pollution complaint. The proper way to deal with these real issues is a cooperative market based solution as will be discussed below.
The employment aspect of AI data centers is of note here; it is a minimal disruption. Data centers do not create a significant number of long-term jobs. This means that centers will not need more traffic lanes or produce ancillary development in the form of support services and retail businesses, the way an office park would. In short, this is a capital-intensive, labor-light industry. This then sets up the scandal of their financing as these centers are being subsidized under laws that were created to incentivize job creation.
Yes, as with most things it comes down to the money and the financing mechanisms. Many centers are funded through a patchwork of tax exemptions, property tax abatements and IRS offered accelerated depreciation schedules. Data centers are also the beneficiary of cheap credit that unsubsidized, non-politically connected businesses cannot get. This puts the government in the business of choosing which sectors to steer capital toward in an explicitly political calculation rather than market-based calculations that will produce better results.
Conclusions
As has been shown, the environmental impact is real, and it is growing. Yet, we have also seen that this impact is comparable to that of heavy industrial uses. In the greater scheme of things this is clearly a manageable impact. This is especially the case when the economic impact of AI is considered. Opponents have only been looking at the costs and have ignored the upsides of what AI (and technology in general) can offer the economy. That will be considered more fully in Part III. Until then, understand that objections are driven more by fear of the unknown than anything grounded in reality. This is a fear that is wholly unbecoming to God’s children. Having enough faith in God to use the brains He gave us should lead us to a dispassionate view of the facts surrounding this (and every other) issue before us.
What we do know should not cause us to fear. There are possible situations whereby a nuisance is created: noise or physical pollution come to mind. This creates the possibility of a tort claim by the aggrieved property owner. Using state power to stop someone from using their own land is illegitimate and a Golden Rule violation. Seeking redress for measurable damage to your property is not. Murray Rothbard dealt with a market solution to these issues. There is a cooperative market process to address these scenarios; property covenants covering setbacks, noise issues and general property usage in a given development. These combined with a legal process to deal with tortfeasors are a far better solution than a coercive governmental ban of data center construction.
It should lead us to recognize that all of this fear-driven consternation is just so much NIMBYism. NIMBYism always has two participants, those who want to proceed with a project and those who want to maintain the status quo. Since all of us are on the YIMBYism (Yes in my back yard) side at some time or other, to fall prey to NIMBYism is to violate the Golden Rule. We can better serve God and our neighbors by embracing consensual cooperation via a nonviolent market rather than falling for coercion as a default setting.
Praise Be to God
