In April–June 2026, local opposition cancelled or delayed at least 45 data center projects in the United States, with a combined planned value of about $68 billion. The latest tally from the research project Data Center Watch also reports that some municipalities have imposed construction moratoriums even before developers file for permits. Demand for AI computing keeps growing, while host communities are being asked who will bear the burden on power and water. New York State's guidance on community investment shows what kinds of commitments may be needed to turn large sums of money into facilities that actually operate.
The $68 billion tally and what it doesn't show
Data Center Watch's public summary for Q2 2026 counts at least 45 projects blocked or delayed by local opposition, with a planned value of about $68 billion. This is not a tally of investment that has ultimately vanished. The figure cannot tell us whether delayed projects will resume under revised terms or move to other sites.
The breakdown available in the public summary is also limited. Project-level data is behind a paid report, and the public portion does not show how many projects were cancelled versus delayed, or how the dollar values were calculated. Nor can a failure rate be calculated, since the total number of planned projects nationwide is not given. The roughly $68 billion should be read as a measure of the scale of projects that faced local opposition.
The permitting developments, however, are concrete. According to the same research, some municipalities have put development moratoriums in place before developers even expressed interest or applied for permits. That means the conditions for building on a candidate site can change earlier than the stage at which individual proposals are reviewed.
States are also responding. Data Center Watch says bills addressing siting, power and water constraints, and the sharing of infrastructure costs were proposed in 30 states, and many were adopted. That does not mean the same regulation has been enacted and enforced in all 30. Bills, resolutions and administrative measures differ in force and scope, so their effect on construction has to be checked locally.
What $68 billion and $31.6 trillion each measure
In its global data center outlook released September 2, PwC projected cumulative capital expenditure of $31.6 trillion from 2026 to 2050 in its central scenario. Set beside news of projects stalling locally, the two may seem contradictory, but their scope and time frames differ.
The $68 billion is the value of US projects cancelled or delayed, the $31.6 trillion is a global cumulative investment forecast that includes equipment replacement, and the $1 million per MW is a voluntary community-investment benchmark set out by New York State. They cannot be compared as like-for-like investment amounts.
| Figure and source | Scope and period | What the figure includes | What it doesn't tell you |
|---|---|---|---|
| About $68 billion / Data Center Watch | US, April–June 2026 | Planned value of at least 45 projects cancelled or delayed by local opposition | Total loss of that amount; a failure rate against all US projects |
| $31.6 trillion / PwC | Global, 2026–2050 | Central forecast of cumulative capex on buildings and structures and on ICT equipment, including replacement | Committed investment; any guarantee that individual projects will be completed |
| $1 million per MW of power demand / New York State | State guidance published September 2026 | Recommended community-investment benchmark, including investment in grid modernization | A uniform tax amount; an annual payment; a guarantee of approval in exchange for payment |
The table compares Data Center Watch's public summary, PwC's forecasting methodology, and Section III of New York State's community investment guidance by scope and the nature of each figure. You cannot divide $68 billion by $31.6 trillion to conclude that opposition has little impact, nor subtract it from the forecast to find the remaining investment.
PwC's forecast includes repeated equipment replacement after facilities are completed. The model, commissioned from Oxford Economics, covers 46 countries and regions and assumes ICT equipment is refreshed every four to six years. Amounts are in real US dollars at 2025 exchange rates. Even if a building lasts a long time, replacing the servers and GPUs inside it raises cumulative spending.
It is also not a sum of announced construction projects. It estimates the power capacity IT equipment will need based on economic growth and digital demand, then converts that into investment through regional supply allocation and unit equipment costs. Depending on how quickly AI spreads, the cumulative figure ranges from roughly $22 trillion to $50 trillion. It is a forecast whose result shifts with how demand is estimated, and it is not evidence that the $31.6 trillion is on track despite continued opposition.
$1 million per MW: New York's terms for returning value to communities
Empire State Development, New York's economic development agency, has published guidance called the Community Investment Framework (CIF) for municipalities hosting data centers. The recommended $1 million per MW of power demand draws attention, but it needs to be read together with the nature of the framework and how the money is used.
The CIF is a voluntary framework, a guide for municipalities and operators negotiating investment in the community. Whether to make an agreement binding is for the municipality to decide with advice from legal experts. Environmental review and required regulatory approvals remain separate. It is not a system in which paying into the community secures permission to build.
Moreover, the $1 million benchmark includes investment in the grid modernization that the state requires. The guidance gives an example on page 6 of $50 million for a 50 MW project, but does not say the full amount would go to the municipality on top of existing grid costs. The initial contribution, and when and how much is paid afterward, are matters for negotiation. MW represents the scale of power required, not a charge based on electricity consumption or an annual amount.
The benefits to host communities also cannot be gauged by total construction cost. The guidance asks that increased property tax revenue be evaluated alongside tax incentives. On construction jobs, it notes that crews traveling in from outside the area may do the work, so communities need to check how much work remains local. Allocating limited land and power to a project also affects opportunities to use them for other industries.
What becomes negotiable, then, is the assets that remain in the community and the cost of maintaining them. The CIF targets community-owned assets such as public roads, and takes into account that facilities built with the funds incur ongoing maintenance costs. It also touches on the possibility of a fund to help address what happens if a data center withdraws, and to support remediation of an abandoned site. If an agreement rests only on the amount received at opening, later expenses could be left with the municipality.
The counterparty needs checking too, since the developer that builds may not be the company that owns and operates for the long term. The guidance advises confirming which entity will ultimately fulfill the commitments and, where both are involved, separating payment schedules to clarify obligations. Whether further community investment will be needed when future expansions raise power demand is also among the items to include in an agreement.
This is a negotiation design set out by New York State, not an obligation applying nationwide. Even so, by going beyond the prospect of higher tax revenue and jobs to spell out who bears which costs and when, it offers material for judging how feasible a project is.
Conditions for turning money into operating capacity
PwC also cites the time needed to secure transmission capacity and substations, and lead times of years for transformers, as constraints on construction. Its view is that policy predictability and community consent are what turn demand into capacity that actually gets built. An announcement of financing alone does not determine when a site can start drawing power.
Moving to a place with less opposition is also not equally easy for every use. PwC explains that AI training is relatively unconstrained by distance from users, while inference is constrained by factors such as response latency, data access, and requirements to handle data domestically. A project stalled in one area cannot necessarily be moved elsewhere for the same purpose.
After the roughly $68 billion in stalled projects, the question to check is not only whether a project has resumed, but what conditions allowed it to. Has the timing of power supply been fixed, have the necessary permits been obtained, and does the agreement with the community spell out maintenance costs and responsibility in case of withdrawal? If those conditions are met, operators can better anticipate the path to operation, and municipalities can decide whether to accept a project with future burdens in view.