seeing across time

Something new appeared in community feeds starting around 2024. Professionally designed infographics. Coordinated websites. Organized social media campaigns mapping proposed data centers to water tables, grid load, tax abatements, aquifer depletion rates, and utility cost projections. The people producing this content are not fringe actors. They are organized, technically literate, and using the same digital infrastructure that data centers exist to power – to resist data centers.

Within a month of launching her crowdsourced reporting platform in April 2026, environmental activist Erin Brockovich had received nearly 4,000 submissions.[1] The most common concern recorded was not noise, not water consumption, not rising electricity bills. It was one word – transparency.[2]

That detail matters more than it appears to. People are not primarily objecting to what data centers do. They are objecting to being excluded from decisions about what gets built in their communities, on what terms, with what obligations attached, and what gets left behind when the capital moves on. The technology they are using to mount this resistance is the same technology that makes data centers necessary. That irony is not decorative. It’s structural, and understanding why requires stepping back from the event and looking at the pattern underneath it.

This is the question worth sitting with – how do we develop discernment in a world where multiple truths are operating simultaneously? The economic development director and the water authority manager and the environmental advocate and the long-term resident are all looking at the same proposed facility. None of them are wrong. They are tracking different patterns operating at different scales, and the institution asked to decide is rarely designed to hold more than one of those patterns at a time.

We Have Seen This Before

The fracking boom offers the closest structural precedent, and the parallel has already been named in print. A Pennsylvania editorial drew the comparison explicitly: two decades after fracking launched in the commonwealth, jobs are minimal, energy prices remain high, health concerns linger, and the environmental impact is ongoing. The language of economic opportunity surrounding both moments sounds, in that editorial's words, distressingly familiar, as does the investment bubble driving each one.[3]

The governing logic is unnervingly familiar. A capital-intensive industry arrives in communities with promises of jobs and tax revenue. The fiscal case is built on projections that assume today's regulatory environment, today's commodity prices, and today's policy framework. In many of the projects critics cite, permanent employment has fallen well below the scale implied by the original economic-development rhetoric, while some infrastructure and environmental costs remain local.

The data center rush differs from fracking in one critical respect, and the difference makes the current moment more consequential, not less. Fracking required specific geology. It concentrated in the Permian Basin, the Marcellus Shale, the Bakken Formation, geographically bounded sacrifice zones whose costs were at least knowable in advance by region. Data centers require power, fiber, land, and water. Those inputs exist almost everywhere. This boom is not arriving in a few places. It is arriving in every market with a highway interchange, a transmission line, and an aquifer. The pattern that produced identifiable sacrifice zones in the fracking era is now operating at a scale that has no geographic limit.

The Scale Most Analyses Miss

The numbers associated with the data center build-out are large enough to be genuinely difficult to appreciate. Goldman Sachs now projects global data-center power consumption rising 175 percent by 2030 compared with 2023 levels, up from its earlier forecast of 165 percent.[4] A Lawrence Berkeley National Laboratory report prepared in response to a congressional directive and released through the Department of Energy, estimates that data centers could account for between 6.7 and 12 percent of total U.S. electricity consumption by 2028, up from 4.4 percent in 2023.[5] Capital commitments approaching one trillion dollars by 2030 represent infrastructure investment at a pace that has few historical precedents outside wartime mobilization.

What those numbers describe is not a trend. A trend can reverse or moderate. What is happening here is a structural commitment – physical infrastructure locked into specific geographies, drawing from specific water systems, loading specific grid segments, whose consequences will persist for decades regardless of what AI does or does not ultimately deliver. The installation phase of a technological revolution builds infrastructure before the productive use of that infrastructure is fully legible. This has happened before. The railroad boom of the 1870s produced more track than the economy required, and the communities that became nodes in that network inherited both the benefits and the stranded costs when the deployment phase arrived differently than the installation phase had suggested. The fiber overbuilding of the late 1990s left physical infrastructure in the ground whose economics took years to resolve. Capital doesn't wait for clarity. It builds first and sorts out productive use later. The communities where it builds hold the consequences either way.

There is a dimension to this that the railroad and fiber analogies don't fully capture. Excess track and dark fiber could be repurposed. A hyperscale data center shell, built for a specific cooling architecture and power draw profile, may have far less adaptive value when the technology changes. Chip architectures are evolving rapidly. Cooling methods are shifting. The AI investment cycle carries real uncertainty about its productive configuration. Communities being asked to structure twenty-year abatements and long-term utility commitments around today's facility design are insuring one side of a technological bet while the capital retains the freedom to exit when the bet resolves.

Against that backdrop, organized opposition blocked or delayed more than $98 billion in data center projects in the second quarter of 2025 alone, more than all previous quarters since tracking began.[6] In the first quarter of 2026, a new record was set: at least 75 projects worth approximately $130 billion blocked or delayed in a single quarter, matching the total disruption for all of 2025.[7] Grassroots opposition organizations had grown to 833 groups across 49 states by March 2026.[7] This is not a local governance problem. It is a signal that something in the pattern is being recognized by people who don't yet have a name for what they're seeing.

The Institutional Architecture of Exclusion

The resistance movements circulating infographics and filing petitions are tracking the mesh. They've identified that a data center connects to water, to grid capacity, to tax structure, to land use, to something that doesn't fit on the project summary. They are right that there is more than meets the eye. What they're also encountering, often without being able to name it, is a governance architecture in which knowledge, authority, accountability, and liability sit in different institutions, and none of those institutions holds the whole picture.

A typical project moves through something like this sequence. A private developer or site selector identifies the location under confidentiality. The ultimate technology company may not yet be publicly identified. A utility negotiates load commitments and infrastructure requirements in a separate process. An economic development corporation or regional partnership structures the initial deal privately, with negotiating capacity but no authority to approve zoning or commit public funds. A local government enters with actual approval authority but often arrives late, after the essential architecture has already been established. The state may be offering incentives or quietly changing utility and tax rules. Regional water authorities, transmission operators, and air quality regulators bear the consequences of the decision without necessarily holding approval authority over it. Residents hold technical public-participation rights and encounter the project only after much of it is already decided.

The transparency problem is not simply that information is hidden, though it often is. The deeper problem is that the actor who knows may not be the actor who approves. The actor who approves may not control the grid or the water system. The institution that negotiated the deal may not be subject to public-records laws. The entity receiving the economic benefit may not bear the long-term public obligation. This fragmentation is not an accident of poor planning. It’s the operational form that mobile capital takes when it negotiates with place-bound institutions.

This governance version of the mesh is what communities are actually navigating, and it generates a set of questions that the permit process was not designed to ask, or answer.

Who knows what. Most communities encounter a data center proposal after the site has been selected, the permits drafted, and the economic development negotiation largely concluded. Research by University of Mary Washington professor Eric Bonds and Viktor Newby examined 31 Virginia localities with existing or proposed data centers and found that 25 of them, 80 percent, had nondisclosure agreements in place.[8] The information asymmetry is not incidental. It is the mechanism through which the most consequential decisions get made before the community enters the room.

Who controls the narrative. The economic development case is structured, resourced, and delivered first. It arrives with renderings, job projections, and tax revenue models. The opposition arrives later, built from public records, crowdsourced data, and whatever a local planning commission could obtain. The sequence shapes what registers as fact and what registers as concern. By the time residents are circulating infographics, the narrative architecture has already been established.

Who controls the information. Tech companies are not required to disclose water consumption data in most jurisdictions. Grid load commitments are negotiated privately with utilities. In many cases, the communities most affected by a facility's resource draw don't learn the actual numbers until a local newspaper files a records request or a resident notices water pressure has dropped. The xAI facility in Memphis had been operating for months before environmental groups using thermal imaging discovered 33 gas turbines on-site emitting an estimated 1,200 to 2,000 tons of nitrogen oxides annually, compounds the company had categorized as too small to require permits.[9]

Who controls the resources. Capital, legal expertise, regulatory relationships, and lobbying capacity are concentrated on one side of these negotiations. Tech giants spent a combined $61.5 million on lobbying in 2024.[10] Communities bring their planning commission and their zoning attorney. The asymmetry is structural, not a reflection of local capacity.

Who gets to participate. Permitting timelines and NDA structures determine when the public enters the conversation. In most cases, meaningful decisions are made before public input is solicited. The hearing is the notification. It is not the deliberation.

Who bears the consequences. Water consumed through cooling does not return to the local system on the timetable at which households, farms, and ecosystems need it. The grid infrastructure built to serve a facility becomes a local cost regardless of whether the facility remains. The tax abatement extends for a decade or more. The permanent jobs are fewer than promised. A 2025 review of more than 1,200 U.S. data centers found that they generally employ fewer than 150 permanent employees per facility, with some employing as few as 25.[11] The community holds all of this while the capital considers its next location.

Who can leave. The developer can abandon a site. The technology company can redirect strategy. Capital can move to another state or another country. A utility may recover infrastructure costs through rates charged to a captive customer base. Residents, aquifers, school districts, and local governments cannot relocate with equivalent ease. This asymmetric mobility is one of the defining structural features of the arrangement, and it rarely appears in the economic development analysis.

What the Rush Actually Costs — and Where

Three scales of harm are operating simultaneously. Each one is real. None of them is fully visible from inside any single permit process, and reading only one produces a distorted picture of what is actually at stake.

At the local scale, the costs become physical and immediate. In water-constrained regions, consumptive cooling demand can compete with household, agricultural, and ecological needs. In places where available grid capacity is already limited, large new loads can require substantial investment in the systems that hospitals, water treatment facilities, businesses, and households depend on. Those pressures become especially consequential in communities already carrying environmental burdens and possessing limited leverage over the terms of new development.

Boxtown, a historically Black neighborhood in South Memphis where reporting based on Census data places median household income at approximately $37,000, sits near xAI’s Colossus facility, which the company described at launch as the world’s largest AI supercomputer. Before xAI arrived, Shelby County had repeatedly received failing grades for ozone pollution from the American Lung Association, while the Memphis area continued to face federal ozone-compliance challenges.[12] xAI initially operated gas turbines at the site without air permits, maintaining that the units were temporary and therefore exempt from requirements that environmental groups argued should apply. Some nearby residents reported persistent gas odors and worsening respiratory symptoms, although no population-level study has established a causal connection.[13] In 2025, the NAACP and the Southern Environmental Law Center issued an intent-to-sue notice alleging Clean Air Act violations at the Memphis facility.[14] The legal conflict later extended to xAI’s Colossus 2 facility across the state line in Southaven, Mississippi, where the groups filed a federal lawsuit in April 2026. The U.S. Department of Justice subsequently asked the court to dismiss that case, invoking federal authority and national-security interests in support of the facility.[15]

The sacrifice-zone pattern is structural. The community already carrying the greatest environmental burden is recruited as infrastructure host because the barriers to entry are lower, the organized opposition is thinner, and the regulatory environment is more permeable. This is the fracking pattern in a new industry.

The grid costs create a second ledger that rarely appears alongside the tax revenue projections. A community may be told it will receive substantial fiscal benefit from a new data center while simultaneously absorbing electricity rate increases through an entirely different system. Dominion Energy in Virginia proposed its first base-rate increase since 1992 in early 2025, driven substantially by infrastructure investment required to serve data center load.[16] The economic development analysis and the utility rate analysis are typically conducted by separate institutions, presented to separate audiences, and never placed side by side. They should be read together, because the community that appears to gain on one ledger may be quietly losing on the other.

At the regional scale, cumulative effects become visible that no individual permit process can see. Two-thirds of all data centers built or in development since 2022 are located in water-stressed areas: southern Arizona, the Colorado River Basin, and Texas.[17] Lake Mead is currently at 32 percent capacity. Lake Powell is at 24 percent. The Bureau of Reclamation has declared a Level 1 Shortage Condition on Lake Mead for 2026, requiring Arizona to cut roughly 18 percent of its annual water apportionment.[18] A 2025 report from Western Resource Advocates estimated that data centers across five western states could consume 7 billion gallons of water annually by 2035, enough to serve the annual needs of up to 194,000 people.[18] Ceres found that planned data-center growth could increase water stress by as much as 32 percent in some affected areas.[19] The original 1922 Colorado River Compact was written for an agricultural economy. The seven states currently renegotiating water rights for a hotter, drier century are doing so without data center demand as a named variable in the framework.[20]

The grid is under parallel stress, and the concern has moved from projected risk to documented operational failure. NERC, North America's electric reliability organization, escalated its alert on data center load risk to Level 3, its highest essential-action tier, in May 2026.[21] The escalation followed a documented series of incidents in 2024 and 2025 during which 1,000 megawatts or more dropped off the bulk power system unexpectedly and without advance notice to grid operators.[22] Summer peak demand is now forecast to grow by 224 gigawatts over the next decade, a more than 69 percent increase over the prior year's forecast.[22] The difficulty is not only aggregate demand. It is the load profile itself. Data centers draw continuous, high-reliability power and can disconnect suddenly and at scale, a behavior the grid was not engineered to manage. When those disconnections occur, the consequences are not abstract. They reach hospitals, water treatment plants, refrigerated supply chains, and homes.

Each of these regional-scale problems is being generated by local-scale decisions that have no mechanism for seeing the cumulative effect. A city approving a permit for one facility is not approving a regional water strategy, but it is participating in one.

The Tax Revenue Argument Across Time

The economic case for data centers rests on a fiscal projection built from today's rules. Tax abatements are modeled against current property tax frameworks. Revenue projections assume current state revenue-sharing arrangements. The ten- or twenty-year horizon of these models is treated as stable because the rules that generate the numbers feel stable.

Iowa school districts made the same assumption.

The SAVE program, Iowa's penny sales tax dedicated to school facilities, equipment, and technology, was considered a stable, long-term funding source. Districts made capital commitments on that basis: bonds, construction contracts, facility renovation plans executed with the assumption that the revenue stream would hold. The Iowa Legislature then redirected a growing share of SAVE funds to property tax relief, increasing the state's cut from 7.1 percent to 25 percent by 2031. More than 100 school districts landed on S&P Global's credit watch list. Cedar Rapids will need a $30 million emergency loan to finish three elementary schools already under construction. The Linn-Mar district was weeks from breaking ground on a $55 million project when the rules changed. "We went through a really in-depth process to uncover the needs of our facilities and developed a plan," said a member of the district's facilities advisory committee. "Unfortunately, local control to make those decisions was taken away."[23]

The fiscal architecture that underpins a twenty-year data center tax abatement is no more durable than SAVE was. One session of the state legislature can redirect the revenue stream a community built its projections around. One change in federal industrial policy can restructure the incentive landscape that made the deal attractive. One shift in how the federal tax code treats data center depreciation can alter the economics of an investment a community was promised would stay. Communities are making long-horizon commitments inside a policy environment with a much shorter horizon than the projections assume.

The pattern here is not that governments are unreliable. The pattern is that decisions made at higher levels of the system land at full force at the local level, with no local recourse and no mechanism for recompense when the equation changes. The community that approved the deal under one set of rules holds the consequences when the rules shift. That asymmetry runs through the NDA structure, through the information gap, through the site selection logic, and through the fiscal architecture. It is the same asymmetry in each case. The form changes. The structure does not.

The Pattern Underneath

People dramatically overestimate events and dramatically underestimate patterns. The data center rush is being experienced as a series of events: proposals, permits, hearings, approvals, each of which can be evaluated, contested, or celebrated on its own terms. But the forces producing those events have been building across multiple systems for decades, and understanding what is actually happening requires seeing the convergence.

The rush is the visible surface of a capital-cycle installation phase, the same dynamic that built too much railroad in the 1870s and too much fiber in the 1990s, arriving simultaneously with a geopolitical race for AI infrastructure supremacy and a communication revolution that requires physical substrate at a scale no prior information technology demanded. These forces are resolving locally in land-use, utility, water, and tax systems designed for a different economy. The communities being recruited as nodes in this emerging mesh are being asked to evaluate a project. What they are actually deciding is which pattern they are joining, on what terms, with what understanding of what changes hands across the time horizon that actually matters.

The resistance movements building crowdsourced databases and circulating infographics are doing something analytically important even without the vocabulary to name it. They have identified that the data center connects to more systems than the permit application acknowledges. They are right. The irony running through all of it, that the technology being used to resist AI infrastructure is the same technology that makes AI infrastructure necessary, is not a paradox. It is the communication revolution demonstrating its own logic. New information architecture always generates both the infrastructure that sustains it and the tools that challenge how that infrastructure gets deployed.

What Discernment Actually Requires

The leaders navigating these decisions most effectively are not the ones who evaluate the project best. They are the ones who can see the pattern the project is part of, who understand that a data center permit is simultaneously a water commitment, a grid commitment, a fiscal commitment extending across a policy environment they cannot fully control, and a statement about which communities in their region will bear the costs of infrastructure that benefits a much wider geography.

That capacity is not a political position. It does not require being for or against data centers. Some communities have welcomed these facilities after genuine public engagement, honest disclosure of resource impacts, and enforceable commitments tied to community outcomes. Microsoft's recent departure from nondisclosure agreements in early-stage community engagement points toward what that process can look like, though the distance between a stated policy and an enforceable practice is itself a pattern worth tracking.[24] No community can opt out of the larger pattern. A community can still choose whether, where, and on what terms it becomes a node within it.

A process capable of making that choice consciously would look different from what most communities currently have. It would require disclosure before site commitments become functionally irreversible. Independent verification of water, power, employment, and fiscal projections rather than reliance on developer-supplied models. Regional assessment of cumulative impacts alongside project-by-project review. Public clarity about who pays for grid and utility upgrades and whether those costs are borne by the facility or socialized across existing ratepayers. Agreements that survive ownership changes and technological obsolescence. Periodic fiscal stress tests rather than a single projection frozen at approval. Enforceable community commitments with clawback provisions and decommissioning obligations backed by financial assurance.

A community does not need certainty before it acts. It needs a process capable of carrying uncertainty honestly.

The fiscal lesson from Iowa is not that tax-based financing is untrustworthy. It is that any financial commitment with a long time horizon needs to be stress-tested against the scenario where the rules change, because across long enough horizons, the rules always do. The water lesson from the American West is not that data centers shouldn't be built in arid regions. It is that water draw is a cumulative regional consequence that no individual permit decision can evaluate on its own, which means the institutions responsible for regional water governance need to be at the table before the facilities are approved. The grid lesson from NERC's Level 3 alert is not that AI infrastructure threatens reliability in the abstract. It is that a load type the grid was not designed to manage requires new frameworks before the load arrives, not emergency protocols after it already has.

None of these insights emerge from evaluating the event. They emerge from seeing the pattern, from holding multiple time horizons simultaneously, from understanding that local decisions participate in regional and national systems operating on their own logic, and from asking not just what a project generates in year one but what the community is still holding in year fifteen.

That is what discernment actually requires in a world where multiple truths are operating simultaneously. Not the ability to choose the right truth from among the competing ones. The ability to see which truths are present, understand how they interact across time, and act with full knowledge of the pattern rather than only the event in front of you.