According to a recent study, the costs of electricity and water are not the only things that AI data centers are negatively impacting
Earlier this month, The Fusion Report covered the concept of building artificial intelligence (AI) data centers in space, with the driving force being the abundance of free solar energy. If you’re following the news recently, you’ve probably heard a lot about the issues with the demand for electricity for these data centers, which are primarily used to support AI workloads (in this article, we will assume that new data centers are predominantly used for AI workloads). You most likely also have read about the need for fresh water to cool data center equipment, and the negative impact this is having on drinking water supplies, especially in places where the demand for water is tight such as Arizona, Texas, and the Southwest. In places like Phoenix, the demand for cooling water for data centers is growing rapidly, yet the area is one that faces severe persistent droughts and rising temperatures. In places such as the Washington DC metro area, particularly in Northern Virginia, water demand has spiked by over 60 percent and data centers now consume roughly 3% of the total water in the Potomac River Basin. Even in places like Utah or Georgia where water availability has historically not been an issue, concerns that are depleted ground water have reached crisis proportion.
And this problem is only likely to get worse, as the demand for data centers continue to grow. Unfortunately, the negative impact of data centers is not limited to electricity and water availability and costs. Recent studies have found that the data center boom has also increased the cost and decreased the availability of construction materials, as well as competing for capital with other uses such as anything other than data centers (think of car or home loans, your company’s recent bond offering, or uses such as commercial fusion energy development). Even creating data center construction jobs has a negative aspect to it, as non-data center construction needs compete directly for construction and trade workers with increased demands from the data center boom.
Data Center vs Non-Data Center Development: The Realities
Historically, construction of manufacturing and commercial facilities results in the creation of jobs for the affected communities. This is not only in the short term, which tend to be construction and trade jobs involved in the building of manufacturing and commercial facilities, but also in the long term. Whether building manufacturing plants, hospitals, hotels, or office buildings, new long-term jobs have historically followed the construction of new facilities. Those new jobs, plus increased municipal and state tax revenues, has been the rationale for granting tax breaks to these projects.
In the case of data centers, the Harvard Gazette has found that the promise of new long-term jobs doesn’t pan out – “It’s a significant false promise of these data centers.” While the construction of data centers requires workers because they are large construction projects, that only generally lasts for a year or two. Data centers that are up and running often require no more than 20 to 50 staff members. If high-tech jobs are created by new data centers, it is often at a distance, not where the data centers necessarily reside. More likely, companies lay off high-tech workers as the number of data centers increase simply to pay for the cost of those data centers and the equipment that goes into them. Even increased tax revenues can be fleeting; one study found that in 2025, Virginia and Georgia have given up more than a billion dollars’ worth of revenue as a result of tax breaks they have provided to attract new data centers. Another study by the University of Michigan found that tax breaks for data centers there exempt companies from paying personal property taxes, including on the billions of dollars of servers and equipment that fill the data centers, tax contributions that have typically helped to fund school systems.
The Impact of Data Centers on Construction and Material Costs
Recent studies have also found that the data center gold rush negatively impacts construction labor costs for skilled construction labor and specialized trades, which can make it harder and more expensive for non-data center projects to hire crews. As contractors shift workers toward higher-paying data center jobs, other projects may face delays, tighter bidding, and higher labor costs. They also add pressure to materials markets by absorbing large volumes of steel, electrical components, cooling equipment, and other inputs, which can lift prices and lengthen lead times for everyone else.
That means non-data center builders often compete with data center projects for the same materials and labor, and they may have to pay more or order earlier to stay on schedule. The effect is usually strongest in markets that have become preferred hubs for data center development, because a small number of mega-projects can absorb a large share of regional capacity at once. In those places, non-data center projects often face tighter bidder pools, longer lead times, and more volatile pricing, while regions with less concentration typically feel less pressure on costs
But Who Really Benefits from Hyperscale Data Centers?
That in a nutshell is the $6.7T question, which is the global capital expenditures (CapEx) forecasted to be spent on data centers and their equipment between 2025 and 2030, with the US spending the largest share of any single country (40%, or $2.68T). Out of the $6.7T, $3.5T will be servers, while $1.6T will be split evenly between storage systems and electrical/ mechanical equipment. The spend on labor, $600B, will be less than 9% of the total. And hyperscale capital investments are not free money; a significant part of it financed through debt. In 2025, hyperscalers has added over $121B in new debt, with $90B of it in the last three months of 2025. UBS analysts believe that the debt borrowing for data centers in 2026 alone could be as high as $900B. If this sounds like a repeat of the 2008 mortgage crisis (where subprime mortgage debt reached between $1.3T and $2.0T), you might be right except debt this isn’t for houses that people live in; it’s for warehouses full of servers.
In a similar manner, the impact of the data center buildout on the stock market, especially tech stocks, is incredibly pronounced. Last Tuesday (June 23rd) there was a sudden wave of selling in major technology stocks, ending a 90-day tech rally, and driving the Nasdaq down by roughly 3% in a single day. That day also saw a plunge in the stock value of SpaceX (NASDAQ: SPCX) to below $150/share (its initial public offering, or IPO value), before climbing back to $156/share at the end of the day. Similarly, Amazon (NASDAQ: AMZN) reached a low of $234.11/share, down from $274.99/share in early June. Other tech stocks such as NVIDIA dropped by 4.15%, Apple dropped 0.91%, and Meta dropped by 0.29% on Tuesday.
Which begs the obvious question: who are the companies that are building these massive data centers and taking on all this debt, with potentially negative implications for their stock prices? The answer wouldn’t be surprising to anyone who knows anything about the data center market. The largest one is Amazon Web Services (AWS), which controls roughly 28%-30% of the world’s global cloud infrastructure; followed by Microsoft Azure which owns 20%-21%; Google Cloud Platform (GCP) which owns 13%-14%; Meta (also known as Facebook); and Alibaba Cloud which owns 6% of the global cloud infrastructure. In total, the “big three” (AWS, Azure, and GCP) alone own 59% of the global hyperscale capacity. And if the companies making money off AI won’t surprise you, the few people making most of the money off AI are even less surprising. These include: Elon Musk, the world’s first trillionaire; Larry Ellison (Oracle), who added $140B to his previously sizeable fortune of $90B just last year; Jeff Bezos, whose assets are worth $244B and owns roughly 9% of Amazon; Larry Page and Sergey Brin of Google are worth $300B and $275B, respectively; Michael Dell of Dell Computers at $230B; Mark Zuckerberg (Meta/Facebook) at $195B; and Jensen Huang (founder of NVIDIA, which powers most AI systems on this planet) at $180B. Notably, you have to get to #9 (Bernard Arnault of LVMH) and #10 (Warren Buffett) before you find people that are not tech billionaires.
Clearly, Fusion Needs AI, But Does AI Really Need Fusion?
There has been a lot written about how AI can speed the race to fusion from plasma stabilization to improving manufacturing techniques to aiding the maintenance of fusion’s complex systems. However, the reverse of this concept (that AI needs fusion) is probably wishful thinking; AI seems to be doing just fine without commercial fusion today. More to the point. AI just might be detrimental to commercial fusion; think about the race to raise capital (and not just R&D capital) as an example. Per Realta Fusion, constructing a 1GW example of their powerplant in production would cost $2B-$3B; if fusion were used to increase today’s global electricity generation by 25% (currently 9,100GW-9,800GW), we would have to build between 2,275 and 2,450 1GW fusion powerplants. Using the Realta Fusion cost estimates (which are some of the more aggressive in fusion), the cost of these plants would be between $4.55T and $7.35T.
That same amount of capital would only finance 130 gigawatt-scale data centers, each estimated (based on a number of studies) to cost roughly $35B, roughly 10x the cost of a gigawatt-scale fusion power plant. And even though fusion’s cost is a relative bargain compared to that of a gigawatt-scale data center, AI clearly gets significantly more investment today than commercial fusion does. While we believe strongly in the value of fusion, convincing the capital markets of this is the challenge that fusion has when competing for capital against AI investment.
Is the AI Data Center Gold Rush Bad for the World’s Economies?
That is clearly an open question today. In a sense, the birth of AI is no different than the birth of Web 2.0, cloud data centers, the software as a service (SaaS) industry, social media, or the internet itself: each of these new technologies changed the economies of businesses that operated within them. Those technology trends eliminated some jobs and created new jobs, but overall they probably had a net positive effect on things (I will leave social media out for now; its value is at best mixed in my opinion). However, AI’s use and justification is fairly different in both scope and magnitude than previous technology trends. Today, AI dominates the global venture capital market, representing approximately 80% of all the venture dollars deployed today. Similarly, it is estimated 330,000 jobs have disappeared since January 2026 because of AI, including companies such as Oracle, Salesforce, Block, Cloudflare, Atlassian, Meta, and Amazon, even though many of these companies have record earnings. Note that neither of these things are a direct result of AI, rather they are a result of human behavior of company executives and investors in many cases justifying choices they would make otherwise, just as they have in previous technology changes. It will probably take several years to sort it all out, but if history is any indicator, greed (which is what AI is really enabling) is usually more bad than good.
