Data Centers Canceled in 2026: The Growing Wave of AI Infrastructure Withdrawals
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you. We only include offers that fit the reading experience.
In the first eight months of 2026, tens of billions of dollars in proposed AI data-center projects were canceled, withdrawn, or formally abandoned across the United States. The largest single case—the QTS Digital Gateway campus in Virginia—was valued near $30 billion. Other multi-billion-dollar proposals in North Carolina, Missouri, Texas, Maryland, Florida, and elsewhere met the same fate. What looked like an unstoppable build-out of AI infrastructure has instead run into a new set of hard limits: local opposition, power timing, water availability, zoning, and permitting risk.
This is not a story of the AI boom ending. Demand for compute remains intense. It is a story of an infrastructure filtering process. Developers and hyperscalers can announce campuses measured in gigawatts and tens of billions of dollars. Turning those announcements into operating facilities is proving far more difficult than the early 2020s suggested.
Part 1 of this series establishes the scale of the 2026 cancellation wave, explains why the issue matters now, lays out the full roadmap for the series, and begins the detailed examination with the single largest withdrawal: QTS Digital Gateway in Prince William County, Virginia.
Why This Topic Matters Right Now
Three forces collided in 2025–2026. First, generative AI and large-scale model training created unprecedented demand for power-dense computing campuses. Second, the physical constraints of the electric grid—interconnection queues, transmission upgrades, and generation timing—could not expand at the same speed. Third, communities that had previously welcomed industrial development began organizing against the noise, water use, land consumption, and rate impacts associated with hyperscale facilities.
The result is visible in the project list below. These are not minor delays. Several represent formal terminations or developer withdrawals of proposals that had already consumed years of planning and political capital.
For investors, utilities, local governments, and technology companies, the practical question is no longer whether AI will need more data centers. The question is which proposed campuses will actually clear the political, electrical, and environmental filters—and on what timeline.
Support independent research — a small pause for something practical:
Full Table of Contents for the Series
- Part 1 (this article): Introduction, why it matters, background concepts, and the QTS Digital Gateway case
- Part 2: Power as the binding constraint — Stream Project Liberty and interconnection reality
- Part 3: Water, cooling, and the Texas and Western examples
- Part 4: Local opposition, zoning, and the rise of formal moratoriums
- Part 5: Mid-sized and smaller withdrawals (New Jersey, New Hampshire, Florida, Maryland)
- Part 6: Financing, lender scrutiny, and the investment implications
- Part 7: What still gets built — the projects clearing every filter
- Part 8: Policy responses and the next 24–36 months
- Part 9–10 (as needed): FAQs, regional deep dives, and forward scenarios
Background: What a Modern AI Data Center Actually Requires
A traditional enterprise data center and a modern AI training campus are different animals. The latter is defined by extreme power density. Individual racks can draw 40–120 kW or more. An entire campus can require hundreds of megawatts to multiple gigawatts of continuous power. Cooling systems—whether air, liquid, or hybrid—consume large volumes of water or require significant energy for mechanical cooling when water is constrained.
Four inputs must align before construction can proceed at scale:
- Land and zoning — large contiguous parcels with industrial or special-use designation, often near existing transmission.
- Power — not just contracted megawatts, but a realistic interconnection timeline and transmission capacity.
- Water or alternative cooling — especially in regions already under drought or aquifer stress.
- Political and community acceptance — local boards, planning commissions, and organized residents can stop or delay projects even when the first three inputs appear available.
When any one of these four is missing or delayed by years, developers face a choice: wait, redesign, move the project, or withdraw. In 2026 a growing number chose withdrawal.
The 2026 Cancellation Snapshot
The table below summarizes the highest-profile cases examined in this series. Values are approximate and drawn from contemporary reporting; exact capital figures can shift with project scope.
| Project | Location | Approx. Value | Primary Driver |
|---|---|---|---|
| QTS Digital Gateway | Prince William Co., VA | ~$30 B | Opposition, litigation, zoning |
| Kingsboro / ESS | Edgecombe Co., NC | ~$19.2 B | Moratorium debate + opposition |
| Missouri AI campus | Pacific, MO | ~$16 B | Local opposition / zoning |
| Stream Project Liberty | Marion Co., SC | ~$800 M | Power timeline |
| Diode Ventures | Henderson Co., TX | Multi-billion | Water + opposition |
| AWS-related Calvert | Calvert Co., MD | Large campus | County moratorium |
| American Tower | New Jersey | ~4 MW | Local pushback |
| Nottingham project | New Hampshire | — | Community uproar |
| Project Jarvis | St. Lucie Co., FL | ~$13.5 B | Opposition / permitting |
These cases are not identical. Some were early-stage proposals that never broke ground. Others had advanced further into the approval process before collapsing. The common thread is that the combination of infrastructure reality and political resistance proved decisive.
Case Study Begins: QTS Digital Gateway, Prince William County, Virginia
The single largest and most closely watched withdrawal of 2026 was the QTS Digital Gateway project. Planned for a roughly 2,100-acre site in Prince William County, the campus was repeatedly described in the $20–30 billion range and positioned as one of the largest data-center developments ever proposed on the East Coast.
QTS, owned by Blackstone, had pursued the project for years. The proposal generated intense local opposition focused on land use, environmental impact, traffic, noise, and the strain on regional power and water resources. Legal challenges and zoning complications mounted. By July 2026 the developer formally terminated the effort.
The significance is not only the dollar figure. Digital Gateway demonstrated that even a well-capitalized owner with deep experience in data-center development could be stopped by the combination of determined local resistance and procedural risk. It also signaled to lenders and other developers that political and legal risk in high-growth data-center markets had risen materially.
Video context: contemporary reporting on the termination of the Digital Gateway project and its implications for Virginia’s data-center corridor. (Replace with the exact high-viewership video ID you prefer from the researched list.)
Prince William County had already become one of the most contested data-center geographies in the country. The Digital Gateway outcome reinforced a broader pattern: jurisdictions that once competed aggressively for data-center tax revenue began imposing stricter conditions, longer review timelines, or outright pauses.
What Part 1 Has Established
The 2026 cancellation wave is real, multi-billion-dollar in scale, and driven by a consistent set of constraints rather than a sudden collapse in AI demand. The QTS Digital Gateway case shows that even flagship projects can be terminated when local opposition, legal risk, and infrastructure questions align against them.
In Part 2 we turn to the power constraint in detail—beginning with Stream Data Centers’ Project Liberty in South Carolina—and examine why interconnection timelines have become a decisive factor for many developers.
Next: Power queues, utility capacity, and the projects that withdrew because the electrons simply would not arrive on time.
[Part 1 Complete. Say "Go" or "Proceed" to generate Part 2.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 2: Power as the Binding Constraint — Why Stream’s Project Liberty and Others Could Not Get the Electrons in Time
In 2026 the most decisive filter on many AI data-center proposals was not land or capital. It was power — specifically, the realistic timeline for delivering hundreds of megawatts of reliable electricity. Stream Data Centers’ Project Liberty in Marion County, South Carolina, became one of the clearest illustrations. The developer was willing. The site existed. The project still withdrew because the electric infrastructure could not be ready on a commercially viable schedule.
Part 1 established the scale of the 2026 cancellation wave and examined the political and legal collapse of QTS Digital Gateway in Virginia. Part 2 turns to the physical constraint that has quietly ended or frozen more projects than almost any other single factor: the gap between announced megawatts and actual interconnection and transmission capacity.
The Power Reality Behind Hyperscale Campuses
A modern AI training campus does not sip electricity. It gulps it. A single large facility can require 200–500 MW or more under full load. Multi-building campuses push into the gigawatt range. That demand must be met with firm, high-quality power delivered through substations, transmission lines, and generation that often do not yet exist at the required scale.
Utilities and regional transmission organizations maintain interconnection queues that have grown dramatically. A project that requests 300 MW or 500 MW may wait years for studies, system upgrades, and construction of new lines or substations. In some regions the queue itself has become a multi-year bottleneck. Even when a utility is willing to serve the load, the physical work of reinforcing the grid can lag far behind a developer’s financing and construction schedule.
This mismatch explains a growing share of the withdrawals recorded in 2026. The AI demand signal is strong. The electrons are not arriving on the same calendar.
A quick, practical pause:
Case Study: Stream Data Centers – Project Liberty, Marion County, South Carolina
Stream Data Centers advanced Project Liberty as a significant campus in Marion County. Contemporary reporting placed the investment in the range of roughly $800 million. The project had local support in some quarters and a developer with experience delivering data-center capacity. What it did not have was a power delivery schedule that matched the commercial window.
South Carolina, like many Southeastern states, has seen rising interest from data-center developers seeking lower land costs and available industrial sites. The limiting factor has repeatedly been the speed at which the utility can reinforce transmission and provide firm capacity. When the interconnection and upgrade timeline stretched beyond what Stream considered viable, the company withdrew.
Project Liberty is instructive precisely because it was not primarily a story of community revolt or water scarcity. It was a story of infrastructure timing. Even a willing developer and a receptive local jurisdiction cannot force electrons to appear years earlier than the grid can deliver them.
Interconnection Queues and the Hidden Timeline Risk
Interconnection is the formal process by which a large new load or generator is studied and approved for connection to the transmission system. For data centers the process includes steady-state, stability, and deliverability studies. If the studies show that the existing system cannot absorb the load without violations, the utility or regional operator requires network upgrades. Those upgrades are often expensive and multi-year.
In practice this produces three common outcomes for data-center proposals:
- Proceed on schedule — rare for the largest campuses in constrained regions.
- Accept a multi-year delay — many projects simply wait, carrying land and development costs.
- Withdraw or relocate — the choice Stream and others made when the delay destroyed project economics or financing assumptions.
Lenders and equity partners have grown more sophisticated about this risk. A term sheet that assumes power in 2027 is far less attractive when the utility’s official view points to 2031. The result is tighter scrutiny of every new large load request and, in some cases, an outright refusal to finance until a clearer power path exists.
On-Site Generation, Behind-the-Meter Solutions, and Their Limits
Some developers have responded by pursuing on-site or dedicated generation — natural-gas plants, battery storage paired with renewables, or even small modular reactors in the longer term. These approaches can reduce dependence on the bulk transmission system, but they introduce new complexities: air permits, fuel supply, noise, additional capital cost, and longer development cycles of their own.
In 2026 the gap between the speed of AI demand and the speed of new firm generation remained wide. Gas turbines face their own supply-chain and permitting timelines. Large battery installations help with peak shaving and reliability but do not create energy; they only time-shift it. Nuclear options remain years away for most commercial data-center use cases.
Consequently, the projects that cleared the power filter in 2026 tended to be those that either:
- Secured early positions in relatively uncongested queues,
- Located next to existing surplus capacity or retiring industrial load, or
- Accepted phased build-outs that matched realistic utility upgrade schedules.
Projects that required simultaneous delivery of several hundred megawatts on aggressive timelines faced the highest withdrawal risk.
Broader Pattern: Power Timeline as a Quiet Killer of Projects
Stream’s Project Liberty is not an isolated anecdote. Across multiple states, developers have cited “power availability” or “interconnection timing” as a primary or contributing reason for withdrawal or indefinite delay. In some cases the language is carefully diplomatic; in others the utility correspondence makes the timeline gap unmistakable.
The effect compounds. When a high-profile project withdraws because power cannot be delivered, other developers and their financiers update their risk models. Jurisdictions that cannot demonstrate credible near-term capacity begin to look less attractive even if land is cheap and local officials are welcoming.
What This Means for the Next Phase of Build-Out
The power constraint does not mean AI data centers will stop being built. It means the surviving projects will be those that solve for electrons early — through location choice, phased demand, dedicated generation, or long-lead utility partnerships. It also means the geographic map of AI infrastructure will be shaped as much by transmission and generation reality as by tax incentives or fiber routes.
In Part 3 we examine the second major physical constraint that has ended or stalled projects: water. Cooling large AI campuses requires substantial water or energy-intensive alternatives. In regions already under water stress, that requirement has proven decisive.
Next: Water consumption, cooling technology, and the projects that ran into drought, aquifer limits, and community resistance to large-scale water use.
[Part 2 Complete. Say "Go" or "Proceed" to generate Part 3.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 3: Water, Cooling, and the Projects That Ran Dry — Texas, the West, and the Limits of Evaporative Cooling
Power is the most discussed constraint on AI data-center growth. Water is the quieter one that has already ended or stalled multiple large proposals. In 2026, projects in Texas and other water-stressed regions discovered that cooling hundreds of megawatts of dense compute requires either large volumes of water or expensive, energy-intensive alternatives. When local aquifers, drought conditions, or community resistance made water unavailable or politically toxic, developers withdrew.
Part 2 examined how interconnection timelines and the simple absence of near-term firm power forced Stream’s Project Liberty and similar proposals off the table. Part 3 turns to the second physical filter: water for cooling, and the real-world consequences when that resource is scarce or contested.
Why AI Campuses Drink So Much Water
Traditional data centers have long used water for cooling, primarily through evaporative cooling towers. AI training campuses amplify the problem. Higher rack densities produce more heat per square foot. Sustained high utilization keeps that heat load nearly constant. The result is a continuous demand for cooling capacity that, in many designs, translates into substantial consumptive water use.
Exact figures vary by climate, cooling technology, and operating profile, but large facilities can require millions of gallons per day under peak conditions when relying on evaporative systems. In regions already managing drought, competing agricultural demand, or declining aquifers, those volumes become a political and regulatory flashpoint.
Closed-loop liquid cooling, immersion cooling, and air-side economizers can reduce or eliminate consumptive water use, but they raise capital cost, complexity, and in some cases energy demand. Not every project can absorb those trade-offs and still meet return thresholds.
Practical tech for everyday reliability:
Case Study: Diode Ventures and the Texas Water Constraint
One of the clearer 2026 examples involved Diode Ventures in Henderson County, Texas. The proposal was multi-billion in scale and attracted attention both for its size and for the water questions it raised. Texas has become a major data-center growth market because of land availability, relatively supportive local governments in many counties, and a competitive power market. Water, however, is not uniformly abundant.
In parts of the state, groundwater districts, surface-water rights, and drought contingency plans create real limits on large new consumptive users. When a data-center campus signals demand measured in millions of gallons per day, local stakeholders — farmers, municipalities, and environmental groups — take notice. In the Diode case, water availability and the political response to projected use contributed to the project’s difficulties and eventual withdrawal trajectory.
Texas is not unique. Similar water-related friction has appeared in Arizona, parts of the Mountain West, and other regions where rapid data-center growth collides with long-standing water scarcity. The pattern is consistent: the first few facilities may clear the process; each additional large campus raises the cumulative impact and the political temperature.
Cooling Technology Choices and Their Trade-offs
Developers facing water limits have several technical paths:
- Traditional evaporative cooling — lowest capital cost in many climates, highest water consumption.
- Hybrid systems — switch between evaporative and dry cooling depending on temperature and water availability.
- Air-cooled or dry coolers — eliminate most consumptive water use, increase energy consumption and capital cost, and lose efficiency in hot climates.
- Direct-to-chip liquid cooling and immersion — dramatically improve heat removal efficiency and can sharply reduce facility-level water needs, but require new server designs, higher upfront investment, and operational expertise that is still scaling.
In 2026 the industry continued moving toward higher adoption of liquid cooling for AI workloads. That shift helps on the water metric, yet it does not erase the problem for projects already designed around older cooling assumptions or located in jurisdictions that simply will not approve large water withdrawals.
Community and Regulatory Pushback on Water
Water fights are rarely abstract. Local residents see competing demands for the same aquifer or river. Agricultural users fear higher pumping costs or declining well levels. Municipalities worry about long-term supply for residential growth. Environmental groups highlight ecological impacts. When a data-center proposal arrives with large projected water use, these constituencies organize quickly.
In several 2026 cases, the water issue became inseparable from broader opposition. Zoning hearings that began with questions about traffic and noise expanded into detailed examinations of water rights, drought plans, and cumulative impact. Some local governments responded with stricter conditions, water-use caps, or outright pauses on new large users until regional water plans could be updated.
The result is a de facto filter: projects that cannot demonstrate a low-water or no-water cooling strategy, or that cannot secure firm water rights without harming existing users, face elevated risk of delay or withdrawal.
Western and Southwestern Examples
Beyond Texas, the Southwest has long balanced rapid growth against limited water. Data-center development in parts of Arizona and neighboring states has already prompted local debates about whether large industrial water users belong in desert basins. In 2026 those debates intensified as AI-driven campus sizes grew. Some proposals advanced with aggressive water-recycling and dry-cooling designs; others stalled when the water math could not be closed to the satisfaction of regulators or the public.
The lesson is portable. Any region that treats water as an afterthought in data-center site selection is now taking on material development risk. The projects that survive are those that engineer water use downward from the first design iteration and engage local water authorities early.
What the Water Constraint Reveals
Water is not a universal blocker. In the wetter Southeast and parts of the Midwest and Northeast, many projects still clear water reviews with manageable conditions. The constraint binds hardest where aridity, rapid growth, and large campus proposals coincide.
Together with the power timeline problem examined in Part 2, water scarcity completes a picture of physical limits that pure capital and demand cannot overcome. The AI infrastructure build-out is being sorted by resource reality as much as by market demand.
In Part 4 we shift from physical resources to political ones: the rise of organized local opposition, formal moratoriums, and the zoning tools communities are using to slow or stop data-center development.
Next: Local opposition, moratoriums, and the political filter that stopped projects even when power and water appeared available.
[Part 3 Complete. Say "Go" or "Proceed" to generate Part 4.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 4: Local Opposition, Zoning, and the Rise of Formal Moratoriums
Even when power and water appear solvable, a third filter has proven decisive in 2026: local political will. Organized community opposition, zoning boards, planning commissions, and county-level moratoriums have stopped or frozen projects that looked viable on paper. The QTS Digital Gateway battle in Virginia was the highest-profile example, but it was far from the only one. From North Carolina and Missouri to Maryland, New Hampshire, and Florida, local governments asserted control over the pace and location of hyperscale development.
Parts 2 and 3 examined the physical constraints of electricity and water. Part 4 turns to the political and legal tools communities are using — and the growing sophistication of opposition groups that now treat large data centers as a land-use and quality-of-life issue rather than an automatic economic win.
From Welcome Mat to Scrutiny
For most of the 2010s and early 2020s, many rural and suburban counties actively recruited data centers. The pitch was straightforward: high capital investment, substantial property-tax revenue, relatively few permanent jobs but high-value construction activity, and a clean industrial profile compared with heavy manufacturing. Tax incentives, fast-track permitting, and supportive resolutions were common.
That consensus frayed as campus sizes grew into the hundreds of megawatts and thousands of acres. Residents began documenting noise from generators and cooling systems, light pollution, construction traffic, concerns about water and power rate impacts, and the transformation of rural landscapes into secured industrial zones. What had been abstract economic development became concrete change in the daily environment of nearby neighborhoods.
A refined break in the reading:
Moratoriums as a Policy Tool
One of the clearest expressions of local control is the temporary moratorium. A county or municipality votes to pause new data-center applications or approvals while it studies impacts, updates zoning ordinances, or revises comprehensive plans. The pause can last months or more than a year.
Calvert County, Maryland, provided a high-visibility example when it moved toward restrictions that affected proposed development near the Calvert Cliffs area, contributing to the withdrawal of a large AWS-related campus concept. Similar pauses and restrictive ordinances appeared or were debated in multiple other jurisdictions during 2026.
Moratoriums do not always kill projects permanently. Some developers wait out the study period and return with revised plans. Others treat the signal as decisive and relocate capital to more receptive counties. In either case the tool changes the risk calculus: time is money, and an uncertain political calendar is difficult to finance.
Case Snapshots of Political and Zoning Friction
Edgecombe County, North Carolina – Kingsboro / Energy Storage Solutions
A roughly $19 billion-scale proposal encountered sustained community opposition and moratorium-related debate. Local concerns about industrialization of rural land, infrastructure strain, and long-term character of the county proved powerful enough to contribute to the project’s collapse.
Pacific, Missouri
A large AI-oriented data-center proposal faced organized local resistance centered on zoning and quality-of-life impacts. Despite the economic development narrative, the political path narrowed until the project was effectively abandoned.
Nottingham, New Hampshire
A smaller but symbolically important case: developer withdrawal followed intense local uproar. The episode illustrated that even modest proposals can be stopped when residents mobilize quickly and municipal boards respond.
St. Lucie County, Florida – Project Jarvis / Sentinel Grove
A multi-billion-dollar campus concept ran into permitting and opposition headwinds. Florida’s rapid growth has produced both pro-development and protectionist constituencies; in this instance the balance tipped against the project.
New Jersey – American Tower
Even a relatively small 4 MW proposal was withdrawn after local pushback, showing that opposition is not limited to gigawatt-scale campuses.
How Opposition Organizes
Opposition in 2026 was more networked and professional than in earlier cycles. Residents formed dedicated groups, hired technical consultants for noise and water analyses, tracked utility filings, and coordinated public-comment campaigns. Social media and local news amplified hearings. In some regions, multi-county alliances shared playbooks.
Key arguments that repeatedly gained traction:
- Noise and low-frequency hum from cooling and generators
- Water consumption in stressed basins
- Visual and light impacts on rural or residential character
- Questions about net fiscal benefit after incentives
- Traffic and road wear during multi-year construction
- Perceived prioritization of out-of-state corporations over local needs
Developers responded with community benefit agreements, sound walls, water recycling commitments, and revised site plans. In many cases those concessions were not enough once trust had eroded or the political cost to local officials had risen.
Statewide and Broader Policy Signals
While most action remained local, 2026 also saw early statewide conversations. New York and other states debated frameworks that could limit or condition data-center growth, particularly where grid or environmental stress was already high. Even the discussion of statewide rules raised the perceived risk for developers evaluating new markets.
Lenders and equity investors took note. Community opposition and moratorium risk moved from footnote to standard due-diligence item. Projects in jurisdictions with recent restrictive votes or active opposition groups faced higher hurdles in the capital markets.
What the Political Filter Means
Local opposition and moratoriums do not halt the entire AI infrastructure build-out. They redirect it. Capital flows toward counties and states that still offer clear, predictable approval paths and toward designs that minimize the externalities residents care about most. Projects that ignore community process or treat local government as a rubber stamp now carry elevated termination risk.
In Part 5 we examine the remaining mid-sized and smaller withdrawals in greater detail and then turn, in later parts, to financing consequences and the projects that are still advancing.
Next: Smaller and mid-scale withdrawals, the cumulative map of blocked capacity, and what the pattern reveals about where AI infrastructure can still be built.
[Part 4 Complete. Say "Go" or "Proceed" to generate Part 5.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 5: Mid-Sized and Smaller Withdrawals — The Cumulative Map of Blocked Capacity
The headline cancellations of 2026 were measured in tens of billions of dollars. The quieter story is the long tail of mid-sized and smaller projects that also disappeared. From a 4 MW proposal in New Jersey to campus concepts in New Hampshire, Maryland, and Florida, the same filters — power timing, water, zoning, and organized local opposition — removed capacity that never reached the national business pages. Taken together, these withdrawals redraw the practical map of where AI infrastructure can still be built at speed.
Parts 1–4 established the major cases and the three primary constraints (power, water, politics). Part 5 inventories the additional withdrawals, shows how they accumulate, and extracts the geographic and strategic lessons for developers, utilities, and investors.
Why the Smaller Projects Matter
A single 4 MW or 50 MW withdrawal does not change national capacity forecasts. Dozens of them do. They also reveal the breadth of the filtering process. Opposition and infrastructure limits are not confined to the largest Blackstone- or hyperscaler-backed campuses. They now appear at nearly every scale once a project requires new zoning, significant interconnection, or visible changes to the local landscape.
For capital allocators the implication is straightforward: diligence that focuses only on flagship projects understates the true volume of blocked or abandoned capacity. The mid-tier and smaller losses are part of the same systemic pressure.
A calm moment in a dense topic:
New Jersey — American Tower Withdrawal
American Tower’s proposed data-center project in New Jersey was modest by hyperscale standards — approximately 4 MW. It still encountered local pushback sufficient to produce a formal withdrawal in 2026. The episode is useful precisely because of its size. It demonstrates that community resistance is no longer reserved for gigawatt campuses. Once residents organize around noise, traffic, land-use change, or perceived industrial intrusion, even smaller facilities can lose political viability.
New Jersey’s dense land-use environment and active municipal boards make it a natural laboratory for this dynamic. The American Tower outcome reinforced the message that developers cannot assume quiet approval simply because the megawatt number looks manageable.
New Hampshire — Nottingham Project
In May 2026 a data-center proposal in Nottingham, New Hampshire, was withdrawn after intense local opposition. Contemporary reporting described community uproar that moved rapidly from online organizing to packed public meetings. The developer stepped back rather than continue through a hostile process.
New Hampshire had not previously been viewed as a primary data-center battleground. The Nottingham case therefore carried outsized signaling value: opposition tactics and resident concerns had become portable. What worked in Virginia or North Carolina could be replicated in smaller New England towns.
Maryland — Calvert County and the AWS-Related Campus
Calvert County’s move toward a data-center moratorium or highly restrictive posture directly affected large-campus planning in the area, including concepts associated with AWS near the Calvert Cliffs region. The combination of local policy action and developer reassessment produced withdrawal of the proposed scale of development.
Maryland’s proximity to the established Northern Virginia data-center market made the Calvert outcome particularly visible. It showed that even jurisdictions adjacent to the country’s densest data-center cluster could decide the costs outweighed the benefits and act accordingly.
Florida — Project Jarvis / Sentinel Grove, St. Lucie County
Project Jarvis (also referenced in connection with Sentinel Grove concepts) represented a multi-billion-dollar campus proposal in St. Lucie County, Florida, valued in reporting near $13.5 billion. Although the earliest public discussion edged into early 2026, the permitting and opposition trajectory kept it relevant to the March–August window of cancellations and withdrawals.
Florida’s rapid population growth has created both strong pro-development constituencies and equally strong protectionist ones, especially in coastal and high-growth counties. In this instance the balance of local process, environmental review, and community input worked against the original campus vision. The project did not advance as proposed.
Additional Mid-Tier and the Broader Pattern
Beyond the named cases, 2026 saw a series of less-publicized withdrawals and indefinite delays in multiple states. Some involved power-timeline gaps of the kind examined in Part 2. Others turned on water or zoning. Still others combined all three. The common feature was that the original development schedule and risk assumptions no longer held.
- Virginia — flagship QTS Digital Gateway
- North Carolina — Kingsboro-scale proposal
- Missouri — Pacific AI campus
- South Carolina — Stream Project Liberty
- Texas — Diode Ventures and related water-constrained concepts
- Maryland — Calvert County restrictions
- New Jersey — American Tower
- New Hampshire — Nottingham
- Florida — Project Jarvis / St. Lucie
The list is illustrative, not exhaustive. Additional smaller proposals in other states followed similar paths.
What the Cumulative Map Reveals
Three strategic conclusions emerge from the full set of 2026 withdrawals:
- No region is automatically safe. Even markets that previously welcomed data centers can shift posture within a single election or planning cycle.
- Scale does not guarantee survival. Multi-billion-dollar projects and 4 MW projects alike have been stopped by the same filters.
- Early political and infrastructure work is now a gating item. Developers who treat community engagement, power studies, and water strategy as late-stage tasks are absorbing elevated termination risk.
The surviving pipeline is concentrating in locations that can still demonstrate credible near-term power, manageable water impact, and durable local political support. Everywhere else, the default assumption has flipped from “likely to proceed” to “must clear multiple independent hurdles.”
In Part 6 we examine the financing and lender response: how community opposition and infrastructure risk are changing the cost and availability of capital for data-center development.
Next: Lender scrutiny, higher risk premiums, and the projects that can still raise multi-billion-dollar financing in the new environment.
[Part 5 Complete. Say "Go" or "Proceed" to generate Part 6.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 6: Financing, Lender Scrutiny, and the Investment Implications of the 2026 Cancellation Wave
When multi-billion-dollar data-center projects are withdrawn because of power timelines, water limits, or local opposition, the effects do not stop at the construction site. They travel into the capital markets. In 2026, lenders, private-equity sponsors, and public investors began applying tighter scrutiny to data-center financings. Community opposition and infrastructure risk moved from secondary considerations to core underwriting factors. The cost and availability of capital for new large-scale AI campuses shifted accordingly.
Parts 1–5 documented the projects that disappeared and the physical and political filters that removed them. Part 6 examines how those real-world outcomes are changing the financing environment for the projects that remain.
From Aggressive Growth Capital to Conditional Capital
In the early phase of the generative-AI build-out, data-center financing benefited from strong demand narratives and a willingness among many lenders to underwrite ambitious development pipelines. Hyperscale credit tenants, long-term leases, and the perception of secular growth supported large construction facilities and permanent take-outs.
By mid-2026 that posture had cooled in measurable ways. Reporting from major financial news outlets noted that lenders were examining community-opposition risk, interconnection schedules, and water availability with greater intensity. Deals that previously might have closed on the strength of a hyperscaler letter of intent now faced additional conditions, higher pricing, or outright delays until clearer path-to-power and path-to-permit evidence was provided.
Something tangible amid the financial discussion:
How Lenders Are Repricing the Risk
Three categories of risk have moved up the underwriting checklist:
- Interconnection and power-delivery risk — Evidence that the utility can deliver the required megawatts on the project’s construction timeline is now table stakes. Soft commitments or multi-year upgrade dependencies are treated as material risks to cash-flow start dates.
- Water and environmental permitting risk — In stressed basins, lenders want clarity on water rights, recycling commitments, and the political durability of those arrangements.
- Community and political risk — Recent local opposition, active litigation, or moratorium votes in the same or adjacent jurisdictions trigger enhanced due diligence. Some credit committees now request explicit political-risk assessments.
The result is not a complete shutdown of data-center lending. It is a bifurcation. Projects with early power positions, low community friction, and strong tenant credit continue to attract competitive capital. Projects that still need to solve one or more of the three filters face higher spreads, tighter covenants, larger equity requirements, or delayed closings.
Impact on Developers and Hyperscalers
For pure-play data-center developers the higher cost of capital and longer diligence cycles compress returns and slow the pace at which new campuses can be launched. Some have responded by concentrating development in proven low-friction markets, partnering more closely with utilities on dedicated generation, or shifting toward smaller, phased projects that match realistic infrastructure timelines.
Hyperscalers themselves retain balance-sheet strength and can self-fund or guarantee many of their own projects. Even so, they are not immune. When a major campus is withdrawn or delayed, the lost capacity must be replaced elsewhere, often at higher all-in cost or on a slower schedule. The internal competition for the next available “clean” sites intensifies.
Public Markets and Equity Implications
Publicly traded data-center REITs, tower companies with data-center exposure, and infrastructure funds have also felt the narrative shift. While long-term demand for AI-ready capacity remains a positive fundamental, the 2026 cancellation wave introduced a new layer of execution risk into growth projections. Projects that once appeared in development pipelines as near-certain contributions to future FFO or EBITDA are now subject to higher probability of delay or removal.
Equity investors have begun differentiating more sharply between platforms with geographically diversified, power-secured pipelines and those more concentrated in contested or constrained markets. The premium attached to “shovel-ready with power” sites has risen relative to early-stage land positions that still require major political or utility work.
What Still Gets Financed
Capital has not left the sector. It has become more selective. The projects that continue to raise large construction facilities and permanent financing in the current environment tend to share several characteristics:
- Clear, contracted or highly probable power delivery within the development window
- Location in jurisdictions without recent restrictive votes or active organized opposition
- Cooling designs that minimize consumptive water use or that have already secured durable water arrangements
- Strong tenant credit (typically hyperscale or investment-grade enterprise)
- Phased delivery that matches realistic infrastructure upgrade schedules
Projects missing one or more of these attributes face a longer road to financial close or must accept less favorable terms.
Longer-Term Investment Implications
The 2026 experience is likely to leave lasting marks on how data-center risk is priced and allocated:
- Higher barrier to entry for new large campuses — The combination of infrastructure lead times and political complexity raises the minimum viable project size and the required development expertise.
- Geographic concentration risk — Capital will continue to favor the subset of markets that still offer speed and predictability, potentially increasing concentration in those locations until new capacity or policy changes open additional regions.
- Premium on integrated solutions — Developers and utilities that can offer power, land, and a workable community path as a package will command stronger pricing power.
- Reassessment of growth narratives — Long-term AI capacity demand remains robust, but the conversion of demand into operating megawatts will be slower and more expensive than many 2023–2024 models assumed.
In Part 7 we turn to the other side of the ledger: the projects that are still advancing, the locations that continue to attract capital, and the design and siting strategies that are succeeding in the new environment.
Next: What still gets built — the campuses, markets, and approaches that are clearing every filter and continuing to expand AI infrastructure capacity.
[Part 6 Complete. Say "Go" or "Proceed" to generate Part 7.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 7: What Still Gets Built — The Projects, Markets, and Strategies Clearing Every Filter
The 2026 cancellation wave removed tens of billions of dollars in proposed capacity. It did not stop the AI infrastructure build-out. While high-profile projects were withdrawn in Virginia, North Carolina, Missouri, Texas, South Carolina, and elsewhere, other campuses continued to advance. The difference was rarely demand. It was the ability to clear the three filters examined in earlier parts: credible near-term power, manageable water impact, and durable local political support. Part 7 examines what is still moving forward and why.
Parts 1–6 documented the projects that failed and the tightening of financing around higher-risk developments. This part turns to the positive side of the ledger: the locations, designs, and approaches that are still converting announcements into construction and eventual operation.
The Common Traits of Surviving Projects
Projects that continued to progress through 2026 tended to share a recognizable set of characteristics. None of these traits guarantees success, but the absence of several of them strongly correlates with delay or withdrawal.
- Early and realistic power positioning — Sites with existing interconnection capacity, completed studies, or clear utility upgrade schedules already underway.
- Low or zero consumptive water designs — Liquid cooling, dry coolers, or hybrid systems that reduce political and regulatory exposure in water-constrained regions.
- Jurisdictions without recent restrictive votes — Counties and states that have not imposed moratoriums or experienced high-profile opposition victories in the prior 12–24 months.
- Phased delivery — Initial phases sized to match available infrastructure, with later phases contingent on additional power or water solutions.
- Proactive community engagement — Benefit agreements, transparency on noise and traffic mitigation, and early outreach that reduces the probability of organized resistance.
A practical pause:
Markets That Retained Momentum
Certain regions continued to attract and advance data-center capital even as others imposed new friction. These markets generally offered some combination of available power headroom, established industrial zoning, prior successful data-center experience, and local governments that still viewed the tax base and construction activity as net positive.
Established corridors with existing transmission infrastructure and a track record of completed projects retained an advantage. Newer markets that could demonstrate surplus capacity or rapid utility responsiveness also remained viable. In contrast, jurisdictions that had just experienced a high-profile cancellation or that had enacted pauses saw capital rotate elsewhere, at least temporarily.
The geographic sorting is not permanent. Policy changes, new generation, or transmission upgrades can reopen previously constrained areas. In 2026, however, the near-term flow of shovel-ready activity concentrated where the path was already clearer.
Design and Technology Adaptations That Help
Technical choices have become part of the political and regulatory strategy. Developers who adopted higher degrees of liquid cooling or dry-cooling systems reduced one of the most emotionally resonant local objections — large-scale water consumption. Those who paired campuses with dedicated or behind-the-meter generation reduced dependence on uncertain utility timelines.
Phasing has also proven effective. Instead of seeking simultaneous approval and power for a full gigawatt-scale campus, some developers advanced a first phase sized to currently available capacity while continuing work on longer-lead infrastructure for later phases. This approach lowers near-term risk and creates operating cash flow that can support subsequent expansion.
The Role of Hyperscalers and Balance-Sheet Strength
Large cloud and AI providers retain structural advantages. Their credit quality supports financing even in a more cautious lending environment. Their ability to self-fund or provide strong contractual backing can keep projects moving when pure third-party development capital hesitates. Their long-term demand visibility also allows them to accept longer infrastructure lead times that would break the economics of a shorter-horizon speculative developer.
That said, even hyperscalers are subject to the same physical and political filters. A campus that cannot obtain power or that faces sustained local resistance still stalls, regardless of the tenant’s balance sheet. The difference is that hyperscalers can often relocate demand to other regions or other projects in their global portfolio more readily than a single-site developer.
What “Clearing Every Filter” Looks Like in Practice
A project that reaches construction in the current environment typically has already answered, with evidence, the questions that caused other proposals to fail:
- When will the required megawatts be deliverable, and what is the contingency if the utility schedule slips?
- What is the consumptive water use under design conditions, and how is it secured without harming existing users?
- What is the local political track record, and what specific mitigation and community commitments are in place?
- How is the project phased so that early capital is not stranded by later infrastructure delays?
Projects that can produce clear, documented answers to these questions continue to attract both development capital and construction financing. Projects that cannot are increasingly filtered out before significant capital is deployed.
Implications for the Next 24–36 Months
The projects that clear every filter today will form the visible bulk of new AI capacity delivered through 2027–2029. Their geographic distribution will shape latency, power-market dynamics, and the next round of local political debates. Regions that successfully host these campuses without major backlash may attract follow-on development; regions that experience new friction will see capital continue to rotate away.
For policymakers the lesson is dual. Jurisdictions that want the tax base and construction activity must offer clearer, faster paths on power and permitting while still addressing legitimate community concerns. Jurisdictions that prioritize other values can effectively limit data-center growth through the tools already demonstrated in 2026 — and should expect developers to respond by going elsewhere.
In Part 8 we examine the policy responses emerging at local, state, and federal levels and the plausible scenarios for the next two to three years of AI infrastructure development.
Next: Policy reactions, possible statewide frameworks, and the scenarios that will determine how much AI capacity actually comes online by the end of the decade.
[Part 7 Complete. Say "Go" or "Proceed" to generate Part 8.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 8: Policy Responses and the Next 24–36 Months of AI Infrastructure
The cancellations and withdrawals of 2026 did not occur in a policy vacuum. Local governments invented or expanded tools — moratoriums, stricter zoning, extended review periods — to assert control. State legislatures began discussing frameworks that could either accelerate or constrain data-center growth. Utilities and grid operators faced new pressure to clarify interconnection timelines. Part 8 examines the emerging policy landscape and the scenarios most likely to shape how much AI capacity actually reaches commercial operation by the end of the decade.
Earlier parts established the scale of blocked projects, the physical constraints of power and water, the force of local opposition, the tightening of financing, and the characteristics of projects that continue to advance. This part looks forward.
Local Policy Tools That Are Now Mainstream
By late 2026 the playbook available to counties and municipalities had expanded and standardized. Temporary moratoriums on new data-center applications give officials time to study cumulative impacts and revise ordinances. Updated zoning can require special-use permits, larger setbacks, noise limits, water-use caps, or mandatory community-benefit agreements. Some jurisdictions have begun treating large data centers as a distinct land-use category rather than generic industrial development.
These tools are politically attractive because they respond directly to constituent concerns while preserving the option to approve projects later under clearer rules. They are also effective: several of the withdrawals examined in this series occurred after or during the application of exactly these measures.
A brief, useful interlude:
State-Level Conversations
Most concrete action in 2026 remained local. At the same time, several states opened broader discussions about whether statewide rules were needed. Topics included grid-impact fees, standardized water reporting for large users, limits on incentives in stressed regions, and possible fast-track processes for projects that meet high environmental or community standards.
New York’s early conversations about data-center policy were among the more visible. Other states with both strong tech sectors and rising local friction began internal reviews. The direction of these discussions is not uniform. Some legislators emphasize protecting ratepayers and water resources; others emphasize retaining economic development and AI-related investment. The net result in the near term is heightened uncertainty rather than a single national template.
Utility and Grid Policy Pressure
Interconnection queues and transmission upgrade timelines sit at the center of the power constraint. In response to the volume of large-load requests, some utilities and regional transmission organizations have begun refining study processes, exploring cluster studies, or signaling the need for earlier developer commitments. Policymakers are watching closely because delayed or canceled data centers still leave the underlying grid-planning questions unresolved.
The tension is structural. AI demand is growing faster than the historical pace of transmission and generation expansion. Policy that simply accelerates every large-load request risks reliability and cost impacts on existing customers. Policy that slows every request risks pushing AI capacity — and the associated economic activity — to other regions or countries. The workable middle path is still being negotiated case by case.
Three Plausible Scenarios for 2027–2029
Scenario 1 — Continued Selective Build-Out (Most Likely Baseline)
Local filters remain active. Power and water constraints persist in many markets. Capital continues to flow to the subset of locations and projects that clear every hurdle. National AI capacity grows, but at a slower rate than 2023–2024 forecasts assumed. Geographic concentration increases in the most receptive regions. Cancellations and withdrawals continue at a lower but still material rate.
Scenario 2 — Policy Acceleration in Willing States
A group of states decides the economic and strategic value of AI infrastructure outweighs local friction. They create streamlined permitting, coordinated utility planning, and possibly state-level siting authority for qualified projects. Capacity additions accelerate in those states while other states remain restrictive. The national map becomes more polarized between “build” and “constrained” jurisdictions.
Scenario 3 — Broader Tightening
High-profile problems (rate impacts, water conflicts, or reliability events) produce wider political backlash. More states adopt restrictive frameworks. Financing conditions tighten further. The conversion of announced AI demand into operating megawatts slows materially. Hyperscalers respond with greater emphasis on efficiency, international capacity, or longer-term generation solutions such as advanced nuclear.
Elements of all three scenarios can coexist. The baseline expectation is that Scenario 1 dominates, with pockets of Scenario 2 in the most motivated states and occasional Scenario 3 dynamics in places that experience acute local conflict.
What Policymakers and Industry Participants Can Still Influence
Several levers remain available:
- Transparent interconnection processes that give developers earlier, more reliable visibility into timelines and costs.
- Water accounting and recycling standards that reduce conflict by making impacts measurable and mitigable.
- Community-benefit frameworks that convert abstract tax revenue into visible local improvements, potentially lowering opposition intensity.
- Coordinated generation and transmission planning that treats large AI loads as a foreseeable driver rather than a series of surprises.
- Differentiated treatment for projects that meet high bars on efficiency, water use, and community engagement versus those that do not.
None of these levers eliminates the underlying tension between rapid AI capacity demand and the slower timescales of physical infrastructure and democratic local control. They can, however, reduce the rate of pure waste — projects that consume years of planning capital only to be withdrawn at the final stages.
In the final parts of this series we collect the most common questions readers ask, summarize the core findings, and offer a concise view of what the 2026 experience means for the longer trajectory of AI infrastructure in the United States.
Next: Frequently asked questions, a consolidated summary of the cancellation wave, and the lasting implications for developers, communities, investors, and policymakers.
[Part 8 Complete. Say "Go" or "Proceed" to generate Part 9.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 9: Frequently Asked Questions, Consolidated Findings, and Lasting Implications
The 2026 data-center cancellation wave raised a consistent set of questions from readers, investors, local officials, and industry participants. Part 9 answers the most common of those questions directly, consolidates the core findings from the full series, and outlines the lasting implications for the next phase of AI infrastructure development in the United States.
Frequently Asked Questions
No. Demand for AI compute remains strong. What changed in 2026 is the conversion rate of announced projects into operating capacity. A meaningful percentage of proposed campuses are now being filtered out by power timelines, water constraints, and local political resistance. The boom is encountering infrastructure and political limits; it is not ending.
Trackers and contemporary reporting in 2026 documented tens of billions of dollars in projects blocked, delayed, or withdrawn. When the largest cases (QTS Digital Gateway near $30 billion, Kingsboro-scale proposals near $19 billion, Missouri near $16 billion, and others) are combined with mid-sized and smaller withdrawals, the cumulative announced value exceeds $100 billion by some measures. Not all of that capacity was shovel-ready; much of it was early- to mid-stage.
The QTS Digital Gateway project in Prince William County, Virginia, stands as the largest and most closely watched withdrawal. Valued near $30 billion and planned for roughly 2,100 acres, it was terminated in July 2026 after sustained local opposition, litigation, and zoning complications.
They were always constraints; the scale of AI campuses made them binding. A single modern AI facility can require hundreds of megawatts and, under evaporative cooling, millions of gallons of water per day. Interconnection queues and transmission upgrades move on multi-year timelines. Aquifers and drought plans cannot always absorb new large users. When project schedules assumed faster infrastructure delivery than the physical world could provide, withdrawals followed.
Some opposition is classic resistance to change. Much of it is grounded in measurable issues: noise, water competition, traffic, visual impact, and questions about net fiscal benefit after incentives. Local governments have legitimate authority over land use. The 2026 record shows that authority being exercised more assertively than in the prior decade.
It will slow the growth rate of domestic AI training and inference capacity relative to the most aggressive earlier forecasts. It will not stop growth. Capacity will continue to come online in the markets and projects that clear the filters. Hyperscalers can also shift some workloads internationally or emphasize efficiency gains. The strategic question is how much of the next wave of frontier AI capability is trained on U.S. soil versus elsewhere.
Treat power, water, and community acceptance as design inputs from the first site-selection decision. Secure realistic interconnection positions early. Adopt low-water cooling strategies. Engage local stakeholders before plans are rigid. Phase projects so early capital is not stranded by later infrastructure delays. The projects still advancing in 2026 are largely those that followed this discipline.
A quiet interlude:
Consolidated Findings from the Series
The 2026 cancellation and withdrawal wave was real, multi-region, and multi-causal. Three filters proved decisive:
- Power timing — Interconnection and transmission upgrade schedules frequently lagged commercial project timelines (Stream Project Liberty and others).
- Water availability and politics — Large consumptive use collided with drought, aquifers, and local resistance (Diode Ventures and Southwestern examples).
- Local political and zoning control — Organized opposition, moratoriums, and planning-board scrutiny stopped or froze projects even when physical resources appeared available (QTS Digital Gateway, Kingsboro, Missouri, Nottingham, Calvert County, and others).
Financing markets responded by applying higher scrutiny and risk premiums to projects that had not yet cleared these filters. Capital did not leave the sector; it became more selective.
At the same time, projects that solved for power, water, and community acceptance early continued to advance. The build-out did not stop; it became more disciplined and more geographically concentrated.
Lasting Implications
For developers and hyperscalers: Site selection and early-stage development must now weight political and infrastructure risk as heavily as land cost and fiber access. The cheapest land in a constrained or hostile jurisdiction is often the most expensive choice once delays and withdrawals are factored in.
For local governments: The tools to control data-center growth are proven. Jurisdictions that want the tax base must still offer predictable processes and address legitimate resident concerns. Jurisdictions that prioritize other values can effectively limit new capacity and should expect developers to respond by relocating.
For utilities and grid operators: Large AI loads are no longer theoretical. Planning processes that provide earlier, more reliable visibility into timelines and upgrade costs will reduce wasted development effort and improve outcomes for both new loads and existing customers.
For investors and lenders: The risk profile of speculative or early-stage data-center development has risen. Platforms with power-secured, low-friction pipelines deserve a relative premium. Projects still solving basic resource and political questions deserve higher risk premia or additional conditions.
For policymakers at the state and federal level: The tension between rapid AI capacity needs and the slower realities of infrastructure and local democracy will not resolve itself. Transparent interconnection rules, water accounting standards, and differentiated treatment for high-standard projects can reduce pure friction without eliminating local control.
Series Roadmap Remaining
A final Part 10 will provide a concise executive summary of the entire series, a short list of indicators to watch through 2027–2028, and closing observations on what the 2026 experience reveals about the collision between exponential digital demand and linear physical systems.
Next: Executive summary, forward indicators, and closing perspective on the AI infrastructure filtering process.
[Part 9 Complete. Say "Go" or "Proceed" to generate Part 10.]
This article may contain affiliate links. If you click and make a purchase, we may earn a commission at no extra cost to you.
Part 10: Executive Summary, Forward Indicators, and Closing Perspective
This final part distills the entire series into a concise executive summary, lists the indicators worth watching through 2027 and 2028, and offers a closing perspective on what the 2026 data-center cancellation wave reveals about the collision between exponential digital demand and linear physical systems.
Executive Summary
Between March and August 2026, a wave of U.S. data-center projects was canceled, withdrawn, or formally abandoned. The largest single case — QTS Digital Gateway in Prince William County, Virginia — was valued near $30 billion. Other multi-billion-dollar proposals in North Carolina, Missouri, Texas, Florida, and elsewhere met similar ends. Smaller projects in New Jersey, New Hampshire, Maryland, and additional states added to the total. Cumulative announced value blocked or withdrawn reached tens of billions of dollars and, by some trackers, exceeded $100 billion.
Three filters proved decisive:
- Power timing — Interconnection queues and transmission upgrades could not match commercial schedules (illustrated by Stream’s Project Liberty).
- Water constraints — Large consumptive cooling demand collided with drought, aquifers, and local resistance (illustrated by Diode Ventures and broader Southwestern cases).
- Local political and zoning control — Organized opposition, moratoriums, and planning processes stopped projects even when physical resources appeared available (QTS, Kingsboro, Missouri, Nottingham, Calvert County, and others).
Financing markets responded with tighter scrutiny and higher risk premiums for projects that had not cleared these filters. Capital did not exit the sector; it became more selective. Projects that solved power, water, and community acceptance early continued to advance. The AI infrastructure build-out did not collapse. It entered a filtering phase.
Final practical pauses in the series:
Indicators to Watch Through 2027–2028
The trajectory of AI infrastructure capacity will be shaped by observable developments in the next 24 months. The following indicators are worth tracking:
- Interconnection queue clearing times in major data-center states — Do average study and upgrade timelines shorten, stabilize, or lengthen?
- New local and state moratoriums or restrictive ordinances — Is the political filter still expanding or beginning to stabilize?
- Water-use reporting and recycling mandates for large industrial users — Are standards becoming more uniform or remaining fragmented?
- Financing terms for greenfield campuses — Are spreads and conditions for higher-risk projects still widening, or is capital becoming more comfortable again?
- Share of new capacity using advanced liquid or dry cooling — Is low-water design moving from differentiator to baseline expectation?
- Geographic concentration of actual construction starts — Are starts clustering more tightly in a smaller set of receptive markets?
- Hyperscaler commentary on domestic versus international capacity — Do major operators signal greater willingness to locate training capacity outside the United States?
- Utility integrated resource plans that explicitly model large AI loads — Are planners treating the demand as structural rather than speculative?
Improvement across several of these indicators would support a faster conversion of demand into operating megawatts. Deterioration would reinforce the selective, slower path observed in 2026.
Closing Perspective
The deepest lesson of the 2026 cancellation wave is not about any single project or any single community. It is about timescales.
Software and model development can iterate in weeks or months. Capital can be raised and allocated in quarters. Land can be optioned relatively quickly. Electricity transmission, firm generation, water rights, and democratic local land-use decisions move on multi-year clocks. When the first set of timescales races ahead of the second, friction is inevitable.
In 2026 that friction became visible as formal withdrawals and canceled campuses. The industry response — more realistic power assumptions, lower-water designs, earlier community engagement, phased delivery, and more selective capital — is the rational adaptation. The policy response is still forming and will determine how much unnecessary friction remains.
AI will continue to require large amounts of computing capacity. The United States will continue to host a substantial share of that capacity. The difference going forward is that the path from announcement to operation now runs through a set of filters that can no longer be treated as secondary. Power, water, and local consent are first-order constraints.
This concludes the 10-part series on data centers canceled in 2026.
The AI infrastructure build-out continues — more selectively, more slowly in contested markets, and more concentrated where power, water, and communities can still be aligned.
End of Series
[Part 10 Complete. Series finished.]
No comments:
Post a Comment