One reinforcing loop visible in the data-center sector, driving AI-capacity growth, interacts with two slower balancing channels — one running through state and local restriction adoption and the other through operator concessions and trust-rebuilding — according to a Brookings tally of at least 15 states weighing moratoriums and at least 100 localities that have approved one. BloombergNEF analyst Caitilín McManus, who tracks power and water for data centers, attributes the friction to pace, observing that “many new players are entering the space, perhaps looking to make a quick buck off the AI boom” and that “when everyone is racing to build as quickly as possible, we really still have to consider what comes at the expense of that speed.” A structural reading of that substrate is that the balancing channels are operating with long delays relative to the reinforcing channel — the configuration most associated with cyclical overshoot.
The reinforcing loop, AI demand → construction rate → operational capacity → AI demand, all edges positive, accounts for the multi-gigawatt announcements the article reports — Oracle’s fuel-cell plan scaled to as much as 2.45 gigawatts at Jupiter, New Mexico; OpenAI’s planned 3.2-gigawatt campus in Effingham County, Georgia; SpaceXAI’s Colossus complex in and around Memphis, Tennessee; the natural-gas plant AWS is financing in Pecos County, Texas. The first balancing loop, construction rate → local grievance → restriction adoption → construction rate, contains both positive and negative edges and runs with a delay on the restriction-adoption edge, since moratoriums and local ordinances take legislative cycles to pass and further cycles to enforce. The second balancing loop, local grievance → operator concessions → community trust → local grievance, also mixes positive and negative edges and runs with a longer delay on the trust-rebuilding edge, because statements such as the OpenAI spokesman’s — “we understand why people are skeptical of data centers, and it’s our responsibility to earn their trust” — describe a construction of credibility over years rather than quarters. The substrate thus supports the description of one tightly coupled reinforcing channel and two slower balancing channels, with the operator-concession channel longer-delay than the restriction-adoption channel — a configuration in which temporary periods of friction are expected before equilibrium reasserts.
Operator site-response decisions, read as multi-criteria choices, fit an additive framework of the SMART family more naturally than an outranking or ideal-point method, because the alternatives differ along a small number of weighted continuous attributes that are not incommensurable. The criteria visible across the article are: permitting speed (high weight under investor and AI-race pressure); moratorium risk (high weight after the Brookings tally); operating-energy and water cost (relevant to AWS energy-and-water lead Brandon Oyer’s stated focus); capital cost (relevant to fuel-cell switching and own-generation builds); and environmental and ratepayer externalities (relevant to Meta’s reported Louisiana bill-reduction claim, to the Trump administration’s reported pressure on operators to fully cover grid-upgrade costs, and to candidates including New Jersey Governor Mikie Sherrill’s reported positioning on utility rates). The Oracle choice at Jupiter, from natural-gas turbines to Bloom Energy fuel cells, scores high on moratorium-risk and externalities and lower on permitting speed; Morgan Stanley’s reported assessment that fuel-cell costs have become more competitive as natural-gas turbines have grown scarcer is the cost-vector development that shifts that option up the ranking. For illustration on an equally weighted attribute set drawn from the article-visible criteria, the sensitivity finding is that the fuel-cell path dominates the gas-turbine alternative once moratorium-risk weight rises above roughly one-third of total weight, while a behind-the-meter gas-turbine build or opaque siting remains preferred where permitting-speed weight exceeds roughly two-thirds. Microsoft’s transparency change — ending non-disclosure agreements that had kept projects out of public view, framed as a five-point “community-first AI infrastructure” program — scores high on moratorium-risk and externalities at low capital cost but at a cost to permitting speed relative to opaque siting; OpenAI’s $80-million community-project commitment in Effingham County scores high on externalities and trust criteria and indeterminate on permitting speed. The ranking is robust in mid-range weight profiles but flips at the extremes, which is why cross-firm heterogeneity of operator choice across sites is consistent with rational multi-criteria optimization rather than evidence of strategic incoherence.
Applied against William Ury’s third-side role set as the organizing scheme, the surrounding community of residents, state legislators, utility regulators, environmental and ratepayer advocates, local journalists, and institutional analysts has filled some roles and left others unfilled. Witness is active: the Brookings tally, WSJ reporting, and McManus’s commentary constitute the public-attention function that gives the friction consequences. Referee is active at the local and state level, where the approximately 100 approved local restrictions and at least 15 state moratoriums under consideration are establishing rules of engagement. Provider is active but uneven: Microsoft’s program, OpenAI’s $80-million commitment, Oracle’s fuel-cell switch, and Meta’s reported bill-reduction pledge all address frustrated needs, but the substrate shows them concentrated among large hyperscalers, with Nebius’s John Sutter’s warning — “there are bad apples in this space, including fly-by-night operators. That is a problem for this industry” — indicating that smaller entrants do not uniformly meet that standard. Equalizer is partially active through Microsoft’s end to non-disclosure agreements; healer is lightly active through the same investments, though as a descriptive point, financial commitments without ongoing relational repair rarely function as healing in the strict sense the role carries.
Several third-side roles appear unfilled in the substrate. Bridge-builder — between hyperscalers and ratepayer or environmental constituencies around grid-cost allocation — appears largely absent; this matters because the Trump administration’s reported position on operator-covered grid upgrades, combined with candidates’ positioning on utility rates, points to an unresolved coalition question that no actor listed in the record is convening. Candidate holders for that role in U.S. infrastructure-governance practice would include state public utility commissions, regional transmission organizations (RTOs), the National Association of Regulatory Utility Commissioners (NARUC), or trade associations such as the Data Center Coalition, none of which appear in the article’s record as convening on the specific sites it names. Mediator at the state-government level — a structured convening function for disputes about specific sites such as Jupiter, Effingham County, Vineland, or Memphis — is also unfilled in the visible record. Arbiter exists in regulatory form but has not yet arbitrated the high-profile sites the article names. Peacekeeper is not yet relevant on the substrate, since no actor has reported a physical-violence dimension; the structural reading would say to watch the grievance-accumulation edge for early signs that this role becomes activated.
The configuration the substrate describes — one AI-demand reinforcing channel pulling one way and two balancing channels pulling the other, with the trust-rebuilding balancing channel the longest-delay of the three — is consistent with the cross-firm heterogeneity the article documents, because under unequal delays the equilibrium reached is a temporary one and rational operators optimize within it differently. None of this argues for any particular intervention. It describes a configuration in which, all else equal, a slower pace of new site announcement relative to the rate of concession delivery would shorten the time the system spends in friction.
Analytical techniques used in this piece
This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.
- Multi-Criteria Decision Analysis
- Scores competing options against several weighted criteria at once.
- Systems Dynamics (Structural)
- Maps a system’s structure — stocks, flows, and the architecture that shapes its behavior.
- The Third Side
- Takes the vantage of the surrounding community that has a stake in resolving a conflict (Ury).