The rapid expansion of data centers in the region brings up big questions—one line of inquiry we haven’t heard discussed as much is what happens to the e-waste and infrastructure of data centers when technology becomes obsolete or a site is shut down?
In this blog, we’ll ask some of these questions, with the goal of finding strategies to ensure that communities aren’t left holding the trash while investors disappear with the cash.
A brief history of data centers
We’ve had data centers since the 1990s and even earlier—as long as we’ve had the internet. However, “traditional” data centers, the kind used for all the internet things we’ve been doing for the last few decades— social media, email, video calls, and more—didn’t need nearly as much space or as much power as generative AI.
And while we’ve had forms of machine learning since 1956, the release of a publicly accessible large language model (LLM) in ChatGPT in 2022 marked a notable change in the footprint of generative AI in both our day to day lives and our community landscapes. When people ask what aspects of the internet we could do without Hyperscale AI Data Centers, I often give the general rule of thumb: “Anything that you could do before 2022.”
Generative AI is just really different from all the other things the internet and “the cloud” of the past. It needs more power and more land. And so, as the hunger and hype for generative AI has grown, so has the need to build more and larger Hyperscale AI Data Centers.
Is there an AI data center “bubble”?
Broadly, an industry bubble is when there’s a sharp surge in the market hype for a specific sector, and that sector suddenly has more excitement and hype than it does actual profits. An industry can’t run on hype and speculation forever; when the profits don’t match the buzz, that’s when you get the market crash.
So that leads us to the question of whether the hype of Generative AI —and the buildout of very expensive hyperscale data centers—can match the profit of Generative AI.
There have been a slew of reports that indicate that the billions of dollars being spent to build hyperscale AI data centers are coming from dubious financing schemes described as “leveraged,” “opaque,” “circular” and other terms that do not inspire my confidence. Many economists have expressed concerns that AI tech firms are using a circular flow of investments that are artificially inflating the value of their stocks.
This creates concerns that profits promised by AI may not match the payments that will be required by these “house of cards” style financial expenditures being used to build data centers. The excellent report ”Bubble or Nothing” by Advait Arun, written for the Center for Public Enterprise in November 2025, details reasons for uncertainty in the AI investment landscape. This March 2026 report from the Vanderbilt Policy Accelerator warns that an AI bubble burst could create an economy-wide crash similar to the 2008 Great Recession. (This podcast and short articles and opinion pieces, Reuters opinion from March cover the same themes.)
The big takeaway here is that the financial models that are currently driving massive data center buildout are not as sustainable as we might hope.
Why does an AI data center “bubble” matter for Appalachia?
At the end of the day, why do I care if we’re in an AI bubble? I don’t have any stocks in any tech companies.
Aside from not wanting to see a repeat of the 2008 recession, I also care very much that Appalachian communities don’t become collateral damage in the financial games that AI companies are playing. I don’t want to see the massive data centers being built today become abandoned brownfields in the next decade.
I spent years working on advocacy for increased funds for abandoned mine land reclamation; I’ve seen firsthand the scars and liabilities left behind when an industry moves on. Appalachia has seen this story too many times: our workers, our resources, and our land have created enormous wealth for industries that have a tendency to disappear when the profits slow down.
Of course, it’s entirely possible that the hyperscale AI data centers being built today will create decades of robust tax base and good jobs. Even then, there are serious questions about e-waste and obsolete tech that need to be asked by any community looking at a potential data center.
What’s in an AI data center, and how much do they cost to build?
If we are wondering what gets left behind when an AI data center closes, the first thing to find out is what’s in them. As we can see from data compiled by Karen Kwok from Reuters Breakingviews using Boston Consulting Group (BCG) metrics, the capital expenditure breakdown of an AI data center shows that the computing hardware is a much bigger part of the financial cost of building a data center than any other aspect.
Multiple articles show similar breakdowns in expenditures, this article from McKinsey, an investment consulting firm, provides similar estimates, their numbers break down as 15 percent for land, materials, and site development, 25 percent for power generation, cooling and electrical equipment and 60 percent for chips and computing hardware.
A small difference in numbers, but the same really big takeaway: the most expensive part of a data center is the computing hardware.
This information makes me wonder: if the chips, GPU and servers are the vast majority of the cost of a facility, does that increase the likelihood of a company abandoning a facility when that computing hardware is obsolete? And if so, how long does the computing hardware last?
How long until AI data center computing hardware becomes e-waste?
Looking through industry materials, it seems that the industry standard is to replace data center equipment every 3-5 years. Those tracking industry trends made note when Microsoft extended the lifespan of its data center equipment from 4 years to 6 years. Newer industry materials note that with planning and maintenance, equipment can last up to ten years.
Even if we make generous estimates and assume that future technology improvements extend this timeline further, it seems that we’re looking at a scenario where in 5-15 years we’ll have facilities full of unusable, obsolete equipment.
Where does all the e-waste go?
First, what exactly is e-waste? Basically, it’s waste from anything that once had a plug or a battery (computers, phones, tablets, printers, microwaves, etc). So, not all e-waste is from data centers.
A 2024 UN Report found that 62 million tons of e-waste was produced in 2022, up 82% from 2010–but that study is looking at all forms of e-waste. That UN Report also found that in 2022, less than one quarter (22.3%) of e-waste mass was properly collected and recycled.
An influx of e-waste isn’t just a logistical question of where to put all that garbage—it’s a potential health hazard. E-waste contains toxic heavy metals and chemicals—such as lead, mercury, cadmium and flame retardants—that can pollute the land and make people sick if not properly handled.
So E-waste is bad, but how much of it is coming from data centers?
A 2024 Nature study warned that generative AI alone could contribute an additional 1.2-5 million tons of e-waste globally by 2030, with about 60% of that coming from the United States. The report also noted that circular economy strategies could reduce e-waste generation by 16–86%.
As the article notes, e-waste contains valuable metals such as gold, silver, platinum, nickel and palladium, if recycled, e-waste could be a potential source for billions of dollars of reclaimed materials. Though recycling e-waste can be a source of critical minerals and other valuable resources, recycling isn’t a perfect solution—it’s dirty and dangerous work. Refurbishing and re-use should also be considered.
This data center industry article highlights a coming e-waste crisis if companies don’t develop more sustainable models for hardware maintenance and replacement. It should be noted that tech companies including Microsoft and Google have announced zero-waste goals; while engagement from these industries will be important to the success of waste reduction measures, enforceable policies mandated by a governing body are a more reliable and consistent approach to meet the scale of this problem.
What other waste materials are a potential concern for data centers?
In addition to e-waste, the equipment used for power generation, cooling, and other functions should be closely monitored, and there should be a plan for how they will be disposed of at the end of their life.
Depending on the power source, data centers may use lead-acid or lithium-ion batteries, diesel fuel and other potential hazards that can become a dangerous liability if facilities are abandoned.
The cooling systems are another potential source of future pollutants. In particular, “Two Phase” cooling systems can include PFAS and other chemicals of concern that are used as a refrigerant coolant when run through copper tubing. These refrigerants must be disposed of either at the end of the cooling system’s life (ten years by some industry materials), or when a data center closes.
One of the big concerns here is that many communities may not even know which of these harmful pollutants may be present in a data center due to non-disclosure agreements and secrecy around proprietary technologies. However, if data centers close down, the communities will be left with these potentially hazardous materials to deal with.
What can we learn from the cryptocurrency bubble?
Something of a parallel can be found in the bitcoin and crypto currency mining industry that had a bubble in approximately 2020-2022. In 2022, Kentucky provided nearly 20% of the collective computing power of the country’s bitcoin mining operations.
To be clear: these crypto mining facilities are NOT similar to data centers as they are showing up today. Crypto-mining facilities tended to be much smaller scale and much more haphazardly built than current hyperscale data center build out. In fact, many crypto-mining facilities were just a series of trailers or shipping containers filled with servers. Hyperscale AI data centers can be four million square feet, the size of more than 20 Walmart Supercenters, including at least 5,000 servers and miles of connection equipment.
However, what these two industries had in common was that the vast majority of their expenses were in their computing hardware. In the case of crypto-currency, many operators chose to abandon these shoddily constructed facilities and move elsewhere when the equipment became obsolete, because it was cheaper to abandon and rebuild elsewhere. Only a few years after they were built, many of these businesses had already moved on, leaving behind abandoned facilities, unpaid debts and e-waste.
Protecting against infrastructure costs
The solution to the potential AI bubble and impending e-waste issues is probably not to cross our fingers. The companies making money off data centers are some of the richest in human history. Our communities should not be forced to pay their bills for them or clean up any mess they leave behind.
One concern that communities should ensure that companies address is the cost of infrastructure improvements or burdens that may be imposed on utilities and communities.
As noted in this Utility Dive article, some approaches to mitigate risk and ensure that data centers cover the costs they impose on utilities and communities—even if the data center closes—include:
- Minimum Demand Charges: Requiring customers to pay for a percentage of their contracted capacity, even if actual usage is lower
- Minimum Contract Duration: Locking in long-term agreements, which may include ramp-up periods as data centers come online
- Exit Fees: Requiring companies to pay hefty fees if they decide to break their contract
- Collateral and Credit Requirements: Mitigating financial risks if data center businesses fail
The takeaway here is that many of the improvements that a data center needs are big, regional expenses. These expenses may make sense if a facility will be contributing to the local economy for a decade or more. But what if the facility doesn’t use as much energy as it asked the utility to build out for their facility, or what if it closes after a few short years?
The minimum contracts noted above would require them to pay for these infrastructure improvements regardless of whether they use them, so local communities aren’t, for example, left on the hook for paying off a water line that was built to a facility that’s already moved on.
The importance of planning for decommissioning
What we learned from the coal industry is that you need a plan to clean up the site from day one – and funding must be set aside while the company is still profitable. The worst time to ask a company to clean up their mess is after they have gone bankrupt and left town.
Decommissioning plans, bonds, etc. are typical with any type of facility that has a limited alternative-use potential, as well as with mining, oil and gas, the nuclear industry, and more. If communities are lucky, they may find another company that wants to take over the site, but the size and complexity of hyperscale AI data centers makes that a shaky proposition. From coal mines to power plants to wind farms, many types of specialized industrial sites make it likely that a community could be left managing the harms and burdens that come with an industrial site if the company becomes insolvent or abandons the site.
A reasonable parallel for this could be found in the renewable energy industry. We know that the expected lifespan of solar and wind projects is typically around 25-30 years. Many communities require companies to have a plan to ensure they won’t abandon the wind farms when the turbines need repair. One possible lever for local communities is to pass ordinances or zoning language requiring decommissioning bonds for certain kinds of new projects, or requiring a decommissioning plan as part of the project approval process. These should include an upfront financial commitment, such as decommission surety bonds.
A wind and solar site is typically much simpler and much smaller than a data center, of course. Data center decommissioning also requires data security measures as well as more specialized hazardous materials handling. While no one solution will apply to all projects, financial guarantees should be in place on all of them.
A typical decommissioning plan should include:
- The developer prepares a Decommissioning Plan prior to the construction of a new project, during the permitting phase, before the project is approved. This plan and budget should be approved by a third party.
- A Decommissioning Plan and bond amounts should be updated periodically over the life of a facility to account for new technologies, inflation, etc.
- The Decommissioning Plan should include a detailed blueprint to return the property to a useful condition, similar to preconstruction condition, or preparation for an alternative reuse
At the end of the day, who should hold the risk if the project fails? The company making the profits off the project, or the local taxpayers?
The role of local policy in protecting communities from an AI data center bubble
Hundreds of communities are facing these questions at the same time. Communities should not be on their own to solve these enormous global questions around data centers. We should also not be afraid that demanding basic respect for our people, our land and our assets will cause companies to go to communities they can more easily exploit.
State and federal policy standards are needed to level the playing field and support strained local leaders to be treated fairly by some of the richest companies in the history of our country.
At the local level, more and more communities are passing ordinances, zoning language, or otherwise requiring standards for bonding and decommissioning plans for data centers, such as those passed in Susquehanna County and Smithfield Township in Pennsylvania and a draft ordinance proposed for Columbus, Oh. An example community ordinance that includes decommissioning language can be found in this Template Data Center Ordinance. More information on passing ordinances can be found in this resource from Young Gifted and Green.
Another option is to negotiate community benefit agreements requiring that data center operators cover, for example, 100% of future decommissioning, site remediation, and e-waste recycling costs. This example CBA from NAACP includes language to that effect.
Local ordinances can be more effective when passed before a data center comes to town; and many ordinances that would require responsible data center development are also useful, common sense approaches to any new development.
Though local approaches are important, local communities should not be left on their own, and many states do not allow local communities to pass protective ordinances like we see in the Pennsylvania examples above.
Good state and national policy is important to create a baseline
We have seen the pattern of companies moving to communities less able to advocate for respect and responsible development, state and federal policy is needed to standardize the rules and make the game more fair for everyone.
The policy landscape around decommissioning bonds for the wind and solar industry, as outlined in this 2024 article from PowerMag and in this 2025 Energy Analytics paper is relevant to thinking about what is possible for data centers, however, that policy is still patchwork and developing.
The policy landscape for protection from stranded assets and from data centers is even more sparse. As noted in this blog from the Sabin Center at Columbia Law, Ohio has already established agreements with data centers with minimal use requirements for electricity, and Wisconsin is considering requiring financial assurances for infrastructure upgrades.
Michigan lawmakers are considering legislation (MI HB 6142) that would prohibit the public service commission from approving a contract for a data center unless it includes a decommissioning plan with financial assurance sufficient to cover the cost of decommissioning a data center, South Carolina introduced a similar bill (S 902) in 2026.
Regarding e-waste, 25 US states currently regulate electronics recycling but these policies may need updating to keep up with the challenges of Generative AI.
In most industries, there are good actors with the intent to do things the right way. Those actors should not have any hesitation with committing to agreements with the local community to responsibly maintain their equipment, clean up after themselves and respect their neighbors.
Join the conversation
This blog is part of a larger conversation on whether responsible date center development is possible that we’re having amidst our network. We have additional data center resources on our data center blog. You can also find our responsible development principles here and a compilation of regional resources on community benefit and responsible development support here. Our paper on the re-use of waste heat from data centers is here, and our monthly data center webinars can be found on youtube, or by topic, below.
August 13: Full Disclosure: Data Centers, NDAs and Transparency
June 4: Data Centers and Construction Employment
May 7: Local Government Resources Regarding Data Centers
April 16: Data Centers 101 and Responsible Development Practices
Feb 5: Catching Heat: The Opportunities and Challenges of Using Waste Heat from Appalachian AI Data Centers
You can find our upcoming data center webinars and events: listed here.