The Balkanization Trap

America’s new AI infrastructure is rising not through coordinated national planning, but via a patchwork of local decisions. From 2020–2025, AI-focused and hyperscale data centers have been sited in virtually every state, yet each project has largely been driven by city councils, county boards, utilities, and state agencies acting in isolation. Thus, the nation’s most strategically consequential compute is being green‑lit (or blocked) by atomized, local processes designed for warehouses and strip malls, not for infrastructure with national‑level externalities.

No federal or unified strategy guides where or how these power-hungry facilities should grow. Instead, development is often reactive – tech companies decide on a location and local authorities scramble to accommodate them, frequently with limited foresight or transparencyvirginiamercury.com 1virginiamercury.com 2. The result is an increasingly balkanized landscape of data centers: a strategic national asset being molded by ad hoc local deals rather than any deliberate national policy.

In Virginia – home to roughly one-third of the world’s data center capacity – the absence of statewide rules has left localities to craft their own responsesvirginiamercury.com 3. A proposed state law to require standardized impact assessments was vetoed in 2025 on grounds it would “limit local discretion”. Governor Glenn Youngkin argued that a “one-size-fits-all approach” would infringe on communities’ autonomy. The veto cemented the status quo: each county or city for itself, grappling with data center proposals without overarching guidance. York County, for example, recently imposed its own standards (requiring proof utilities can handle a center’s load), while Loudoun County – the heart of “Data Center Alley” – revoked by-right zoning to force public hearings for new projects. Elsewhere, Henrico County fought internally over basic siting rules. This local-rule patchwork means that approval processes and requirements vary wildly from one jurisdiction to the next, often favoring speed over scrutiny. Tech firms can “forum-shop” for the friendliest regulations, and some counties effectively compete to offer the smoothest ride, trading expedited approvals or tax incentives for promised investment.

Crucially, utilities and regulators are similarly fragmented. A utility like Idaho Power, for instance, must negotiate special contracts to serve massive new data center loads on its grid – essentially making policy on the fly for its service area[9][10]. When Meta and a Google-linked project both chose sites in Kuna, Idaho, the local utility scrambled to arrange new power generation and transmission, all through case-by-case deals reviewed by the Idaho Public Utilities Commission[9]dailyenergyinsider.com 4. There was no national or regional strategy dictating whether it made sense to place two huge AI data centers in that community; the decisions were driven entirely by the companies’ siting choices and the utility’s reactive accommodation. Regulators’ role was limited to ensuring the utility’s existing customers wouldn’t directly subsidize the newcomers – a narrow focus that doesn’t address broader strategic concerns (like regional grid strain or national security implications of clustering critical facilities).

From a national perspective, this fragmented governance yields an opaque and misaligned build-out. Federal officials have heralded the economic promise of a data center boom – at one point even touting a prospective \$500 billion private investment in new U.S. data centerswbhm.org 5 – but in practice Washington provides little coordination beyond cheerleading. Indeed, the only notable federal initiative has been a push for deregulation and speed. In late 2024, President Trump’s “AI Action Plan” called for streamlining permits and even suggested locating data centers on federal land to bypass local hurdlestechtarget.com 6. Yet this top-down nudge toward faster construction does nothing to reconcile where centers should go or how benefits and burdens are managed. If anything, it clashes with the groundswell of local concern: communities from Arizona to Virginia are protesting the impacts of data centers in their backyards, throwing up roadblocks that a federal fast-track policy didn’t anticipatevirginiamercury.com 7. The result is strategic disarray. The nation is racing to build the physical backbone of AI computing, but doing so in a chaotic, decentralized manner – one that risks undermining long-term priorities like grid reliability, resource sustainability, and equitable growth.

The Opaque Playbook of Data Center Siting

Compounding the fragmentation is a culture of secrecy in how major tech firms execute their site selection. Across the country, hyperscale data center projects often arrive shrouded in code names, shell companies, and non-disclosure agreements – tactics that exploit the siloed nature of local governance. With no requirement to disclose plans at a national level, companies operate in stealth until the last possible moment, limiting public scrutiny or coordinated oversight.

Non-Disclosure Agreements (NDAs) have become a standard opening move. It is common for corporations and local officials to sign NDAs at the earliest stage of negotiationstechpolicy.press 8publicrecordmedia.org 9. In Sherburne County, Minnesota, for example, Google entered secret talks with officials under an NDA well before any public announcement. County and city leaders agreed in 2018 not to reveal the identity behind “Project Pine,” as the proposed data center was code-named. For over a year, they quietly assembled tax breaks and infrastructure plans for Google while the public was kept entirely in the dark. Emails later obtained through public records show local economic developers were already discussing “Draft Abatement Agreement” terms for Project Pine even as they were prohibited from saying the word “Google” in public. The pattern repeats nationwide. In Tucson, Arizona, officials spent two years planning a massive 290-acre data center project under NDAs that gagged them from alerting the communityazluminaria.org 10. Pima County supervisors admitted it was “challenging and shocking” to navigate such a consequential decision with an NDA in place. One county supervisor questioned why “protecting the interests” of a private company “supersedes our obligation” to the public’s long-term interests. By the time details leaked – in this case, investigative reporters exposed that “Project Blue” was an Amazon Web Services (AWS) plan – local leaders and a developer had already inked a land deal. The community was essentially presented with a fait accompli, having been denied a chance to weigh in during the formative stages.

Code names are a key companion to NDAs. Major projects seldom carry the tech giant’s name in early paperwork; instead, they go by opaque monikers. In Mississippi, officials referred to AWS’s \$10 billion investment simply as “Project Atlas” during legislative approvalssiteselection.com 11. Only after Governor Tate Reeves signed the incentive package did Amazon’s identity become public, effectively confirming the obvious once secrecy was no longer useful. Likewise, Amazon’s ill-fated plan in Tucson was known only as “Project Blue” in local documentsazluminaria.org 12, and Google’s Minnesota project operated as “Project Pine” until its official unveilingpublicrecordmedia.org 13. These code names are not trivial niceties – they are deliberate devices that limit transparency. By concealing which company is behind a proposal, they hamper the ability of residents, competitors, or even state regulators to fully evaluate the project early on. A bland alias reveals nothing about the proponent’s track record or resource demands, preventing questions that might otherwise arise (for instance, a project called “Pine” raises fewer red flags than one explicitly tied to a company known to guzzle water or power in its other data centers). “It is, in their defense, common,” notes one industry observer, “a lot of economic development projects start with NDAs [and aliases]”, but at the scale of hyperscale data centers the effect is an opaque process invisible to those not in the roomtechpolicy.press 14.

Perhaps the most cunning tool is the use of shell companies and LLCs to mask corporate identities in land deals and permit filings. Tech giants rarely buy land or file permit applications under their household names. Instead, they deploy subsidiary LLCs with nondescript names to operate in stealth. In Idaho, Meta Platforms’ entire \$800 million data center project was fronted by Brisbie, LLC, a Delaware-registered entity that gave no hint of its connection to Meta/Facebookboisedev.com 15. When Idaho Power submitted a required service agreement to state regulators for the Kuna data center, it listed the customer as “Brisbie LLC” – only investigative probing revealed this LLC was owned via layers of shells tracing back to Meta’s headquarters. Google employed a similar tactic in Minnesota, using Jet Stream, LLC (a wholly-owned Google subsidiary) as the signatory to NDAs and purchaser of land for its Becker, MN sitepublicrecordmedia.org 16. Jet Stream’s involvement allowed local negotiations to proceed without tipping off the public that one of the world’s largest companies was behind the project. It was only when the project hit the state Public Utilities Commission (PUC) – triggering public filings – that “the Eagle has landed” email went out, meaning the secret was finally outed. By then, Google’s plans for a \$600 million center (and local tax abatements to subsidize it) were essentially a done deal. This shell-company ruse is widespread. Meta’s Brisbie LLC, Google’s Jet Stream LLC, Microsoft’s frequent use of third-party developers, and Amazon’s habit of transacting via aliases (often under its “Vadata” subsidiary) all exemplify an intent to hide the hand of the true player until late in the game.

The cumulative effect of NDAs, code names, and shell entities is a development process that’s opaque by design. Local officials often justify the secrecy as necessary to attract investment – a “standard due diligence” as AWS put it when pressed about Project Blueazluminaria.org 17. There is some truth that companies demand confidentiality to prevent rivals from swooping in or land prices from spiking. However, this secrecy also means local stakeholders and the public are sidelined until decisions are all but made. Even statutory public hearings, when they finally occur, can be superficial. In the Tucson case, community members blasted the process for lack of “transparency and accountability,” saying, “We have no idea who’s behind all this”. At the Pima County land sale meeting, every speaker given ample time was a paid representative in favor of the project – while residents opposing it were limited to two-minute statements at the microphonetechpolicy.press 18. The information imbalance is stark: Boards hear rosy presentations on jobs and investment from companies and their proxies, with the company name and critical details often still withheld, then are asked to fast-track approvals before public opposition can mount. In many cases, even other local agencies are only partially informed. For example, when Diode Ventures negotiated mitigation payments for its huge data center campus in Idaho, the Kuna school district “didn’t understand the size and scope of the project” and accepted a token sum, only later realizing how under-resourced their agreement wasboisedev.com 19. Such selective disclosure – telling each stakeholder only what is needed to get a green light – exemplifies how fragmented governance can be gamed. Table 1 summarizes some of these prevalent siting tactics and illustrates how frequently they have surfaced in projects around the country.

Table 1. Common Siting Tactics in Data Center Projects (2020–2025)