Source: S&P Global Media Portal via S&P Global.
While capital expenditure for AI data centers continues to increase year over year, there also seems to be an increase in delays and cancellations of new projects. These setbacks are due to various factors, including public resistance, access to power, onerous permitting pathways, and newly introduced legislation that may increase costs or revoke tax exemptions. This shift marks the beginning of a market transition from an era of “expansion at any cost” to one of portfolio optimization, in which developers increasingly treat projects as options that can be augmented, repositioned, paused or abandoned as constraints tighten and pushback increases.
The Take
Local governance and community acceptance have become binding constraints. Anti-data-center sentiment has gone viral, and local pushback has spread even to markets with no data center footprint. In turn, county committees have become more critical of incoming projects, often rejecting them in the early stages. Regardless of the specific pushback, the root of the issue stems from misunderstanding and mistrust. Industry evangelists often outline how much of the outcry is based on extreme outliers or outdated numbers, while current project data is shrouded in nondisclosure agreements. Regardless of the fallacious nature of the adage, the thought that “If you have nothing to hide, you have nothing to fear,” remains persistent among skeptics. Microsoft Corp. recently became the first of the tech giants to announce an end to NDAs with local governments; however, it will take a more coordinated effort to reverse widespread mistrust.
Context
To gain a better understanding of the scope of the issue, we began tracking delays and cancellations in the early stages of data center development. The map below provides a visual of status outcomes arising from a wide range of situations, including withdrawals, rejections and meeting pushbacks. This has been condensed to either canceled or delayed. Projects are considered canceled if the operator has expressed that it will not move forward in that location, or the legislative body has expressed that the project will no longer be considered. Delayed is applied when the operator has expressed that the project will go forward at a later date or that the proposal will be revisited. A third status, delayed*, means that it is unknown whether the operator will revisit. This collection of delayed or canceled projects is focused on early-stage projects that have seen delays prior to the completion of the approval and permitting process, as these are most associated with the recent influx of public pushback and legislation.
451 Research by S&P Global is tracking 47 contested/delayed/canceled projects this year and 103 total since January 2025. So far, 44 have been fully canceled/withdrawn — about half of those in 2025 and half so far this year, so there is an uptick in both early-stage delays and fully canceled projects year over year.
About 78% of the total preapproval delays and cancellations are explicitly tied to public pushback in the municipality. Only about 9% of the tracked projects attribute legislative action to their decision. While reasons for public pushback are myriad, the two most common concerns involve water and power use.
451 Research is tracking about 1,520 land-held and planned data centers across the US. This translates to less than 7% of data centers facing significant delays or cancellations attributable to local pushback in the US.
Water use
One of the major concerns of opposition groups is the water use of data centers. Water is used in data centers primarily through external cooling towers that evaporate water to dissipate heat, a process that consumes significant amounts of water, especially in evaporative systems. Alternatives like compressor-based cooling reduce water use but increase energy consumption, creating a tradeoff that depends on local resource availability. Today, most data center designs use a closed-loop compressor-based system that uses significantly less water than evaporative systems.
Compared with some industries, data center water consumption is lower than public debate often suggests. For example, Google self-reportedly consumed 7.7 billion gallons in 2024, which the company compares with the annual water use of about 51 golf courses. That may sound large in isolation, but Florida alone has more than 1,200 golf courses. All this underscores how limited hyperscaler water use is relative to other familiar land uses. Agriculture is an even bigger water consumer. US corn production alone uses more than 150 billion gallons of water annually, much of which is associated with livestock feed and ethanol, dwarfing the totals reported by major data center operators.
A better place to leverage water concerns in regard to data centers is power generation. In the United States, data centers consume about 1.8 liters of water per kilowatt-hour of electricity used. By contrast, the weighted national average for power production is 7.6 liters per kilowatt-hour across thermoelectric and hydroelectric generation. Hydroelectric facilities can consume as much as 68 liters per kilowatt-hour through reservoir evaporation, although they represent a relatively small portion of US generation. Nuclear, often promoted as a clean solution for digital infrastructure, consumes roughly 1.5 to 3 liters per kilowatt-hour and, across the US fleet, uses more than 300 billion gallons of water annually. There is no doubt that data centers use a significant amount of water, but it is worth noting that in the context of other water users.
Power rates
Another point of common concern is that data center expansion will increase utility costs. Retail electricity prices in the US are increasing at a greater rate than inflation. There is also a nationwide push to insulate retail customers from the financial impacts of data center expansion.
Overall, the issue is nuanced and reveals a double-edged sword. Data center operators are generally responsible for paying for the generation and transmission infrastructure associated with their facilities. Issues arise when a new influx of supply stresses the transmission infrastructure of the wider region, leading to necessary transmission upgrades that are not directly related to the data center. As a whole, the US grid is outdated. Many of these upgrades arguably should have been made years ago, but the projects were not profitable to utilities or would have raised rates with little perceived value to ratepayers and stakeholders.
Therein lies the dilemma: Someone must pay for these upgrades. It is also the case that many of these upgrades are necessary regardless of the increase in capacity. At the same time, the upgrades have now become impossible to ignore due to the influx of data center capacity demands.
Legislation
Public sentiment and lawmakers have made it clear that data center operators ought to be responsible for footing the bill. The current US administration has stated that data centers should be paying their fair share so that ratepayers are not affected. It was later announced that Google, Microsoft Corp., Meta Platforms Inc., Oracle Corp., xAI, OpenAI and Amazon.com Inc. signed the “Ratepayer Protection Pledge,” an ultimately toothless pact to build, bring or buy new power to support data centers in order to prevent higher electricity costs for consumers.
California, Colorado, Georgia, Kentucky, North Carolina, South Carolina, Tennessee Wisconsin, and others have introduced bills aimed at ensuring rate payers are not affected by an influx of data center demand. Florida, Oklahoma, South Dakota, and others have formally enacted a version of ratepayer protection from data centers into law. Maine passed a moratorium to study the issue and create standards, but the bill was vetoed. At the city/county level, these moratoriums are more numerous and growing.
There is also a broader increase in anti-data-center legislation across the US. Lawmakers in at least 23 states have introduced some form of restrictive legislation in 2026 alone. Lawmakers from 12 states have either proposed or announced their intent to propose statewide moratoriums on new data center projects. The overall sentiment from the states seems to say, “Let’s hold off on this for a moment so we can take stock and create standards to ensure mutual beneficence for all parties.”
Moving forward
Historically, early stages of data center projects were largely treated as an oversight. The concern for delays was situated in later stages — after rezoning, approval and permitting. This is no longer the case; due to the negative spotlight on data centers, the initial stages can be fraught with setbacks. Projects in some locations are not even certain once approvals have been granted, as there have been numerous examples of lawsuits from locals or environmental groups further delaying or even overturning approval decisions. Moving forward, it is important to consider where a project is situated in the approval process. The announcement of a project and the often-optimistic timelines have little bearing on project outcomes. It is also important to consider that pipeline numbers, including projects yet to receive approval, are likely inflated. This is becoming more evident to utilities and ISOs as there are increasing specifications of “late-stage” project pipelines, although it is important to understand what “late-stage” means for each group.
The space economy’s next big bet: Orbital data centers
Want insights on datacenter trends delivered to your inbox? Join the 451 Alliance.
