AI Infrastructure as Public Utility: The Growing Debate Over Regulating Data Centers and Chip Supply Chains

A Virginia household's monthly electricity bill jumped from roughly $100 to $281 this January. Nearly three-quarters of the state's voters blame data centers. Meanwhile, President Trump signed an executive order on June 2 restricting access to the nation's most advanced AI systems to roughly twenty vetted organizations—the same month that communities across America discovered they have no say over the data centers claiming their power grids, no voice in chip export policies determining which regions get to build AI industries, and no seat at the table where the costs and benefits are being divided up. The federal government calls this national security. Local officials call it taxation without representation.

AI data centers are projected to nearly double U.S. electricity demand to 150 gigawatts by 2028, equivalent to adding Spain's entire energy needs in just three years. Memory chip shortages have become a new bottleneck for AI infrastructure, driving up costs for consumer devices as manufacturers redirect limited memory capacity toward lucrative AI data center workloads. While Washington frames AI as a national security concern, the urgent civic question being sidelined is whether communities get democratic input on data centers claiming local resources and chip shortages determining economic winners and losers.

Washington Locks the Door

Executive Order 14409, signed June 2, 2026, established a voluntary pre-release oversight framework for frontier AI models and required federal agencies to strengthen cybersecurity, marking a shift toward treating AI as a national security asset. OpenAI launched GPT-5.6 Sol, Terra, and Luna models in a tightly controlled limited preview, with access restricted to approximately twenty U.S. government-approved or vetted organizations—the first U.S. frontier model launched under a government-managed access list.

OpenAI acknowledged: "This government-gated rollout is not ideal or sustainable and broader availability is expected in the coming weeks."

Anthropic accused Chinese tech giant Alibaba of conducting a large-scale "distillation" campaign using 25,000 fake accounts and 28.8 million interactions to illicitly extract capabilities from its Claude AI models between April 22 and June 5, 2026. Anthropic described it as "brazen and unlawful extraction of Claude's capabilities via nearly 29 million interactions using thousands of deceptive accounts."

This national-security framing by federal gatekeepers contrasts sharply with the question of whether AI infrastructure should be regulated as public-interest infrastructure with community input.

Chip Shortages, Economic Winners and Losers

The main chokepoints in AI chip supply chains have moved downstream from wafer fabrication to advanced packaging and high-bandwidth memory, both under strain from surging AI data center demand and subject to escalating U.S. export controls. IDC projects a 13% decline in global smartphone shipments in 2026 as memory is increasingly diverted from consumer electronics to AI infrastructure.

Export controls on AI chips leave 120 countries facing semiconductor limits—including U.S. partners—creating market distortions that strand developing countries' industries or force reliance on foreign alternatives. U.S. AI chip export controls primarily create short-term economic losers for U.S. firms like Nvidia while potentially accelerating long-term technological self-sufficiency in China, with the outcome depending on whether AI breakthroughs occur in the short term or take a decade or more.

OpenAI and Broadcom unveiled Jalapeño, a custom inference ASIC designed solely for large language models, developed in nine months with AI-assisted tools and scheduled for deployment in late 2026 in gigawatt-scale data centers.

These supply-chain decisions—who gets chips, who gets memory, who gets locked out—are being made by federal export authorities and corporate actors with no public deliberation about economic equity or regional access.

Your Grid, Their Power

AI-optimized data center racks demand 30 kW to over 100 kW compared to 5–15 kW for traditional racks, overwhelming local substations. By 2028, data centers could consume 12% of all U.S. electricity. Two planned data centers in Wisconsin by Microsoft and another company will require a combined 3.9 gigawatts—a demand equivalent to 4.3 million homes.

Wholesale electricity costs near U.S. data centers have risen by 267% over the past five years. AI data centers contributed a $9.3 billion price increase in the PJM electricity market's 2025–26 capacity pricing. The Joint Legislative Audit and Review Commission projected that data center growth could drive Dominion residential bills up by $444 per year by 2040. Connecting new data center campuses to the grid faces delays of over three years, creating a critical structural bottleneck. By 2026, global data center energy consumption could approach 1,050 TWh, making data centers the fifth-largest energy consumer in the world if treated as a country, between Japan and Russia.

Yet the debate over whether AI data centers should be regulated as public-interest infrastructure—with community input on siting, grid access, and cost allocation—is being bypassed in favor of closed-door federal permitting acceleration and corporate deals.

Three Decisions, No Public Debate

A June 23 analysis outlines three strategic decisions facing governments by the end of June 2026: whether frontier AI access will be shared via trusted-partner systems or reserved as a national-security privilege, how export controls and chip-market dynamics will shape who can build advanced AI, and whether AI data centers will be regulated as public-interest infrastructure. Model access, chip controls, and infrastructure rules are intertwined and collectively determine who benefits from, and who bears the costs of, advanced AI systems.

The first decision—model access—is being made through executive orders and government-managed access lists. The second—chip export controls—leaves 120 countries facing limits with no public process for weighing economic equity against security claims. The third—infrastructure regulation—remains unresolved at the federal level: the EPA announced it will not pursue nationwide environmental requirements for AI data centers, instead leaving best practices to states and local communities, shifting regulatory responsibility to a fragmented landscape.

None of these decisions involves public deliberation, community hearings, or democratic accountability mechanisms.

States Scramble to Protect Themselves

At least 27 states have advanced legislation to regulate AI data centers as large-load facilities, pivoting from unconditional incentives to frameworks requiring data centers to pay for grid infrastructure upgrades their facilities necessitate. At least 18 states have introduced bills creating special rate classes or infrastructure cost-sharing mandates for large energy users. The principle that data centers must fund their full share of generation, transmission, substation, and distribution costs is being solidified in state regulations, ensuring they do not burden ratepayers for infrastructure not required "but for the new large load."

Ohio's Public Utilities Commission ruled that tech companies must pay special tariffs for data center electricity to cover grid upgrades, while Oregon's POWER Act directs the Oregon PUC to create a separate rate class for facilities using 20 MW or more. Idaho's House Bill 911, effective July 2026, requires any new load above 50 MW to receive service only under a PUC-approved contract that satisfies a "no harm" test, while Texas S.B. 6 regulates large-load customers with demand of 75 MW or more.

Virginia passed 15 data center bills in 2026, including measures on energy cost shifts, water reporting, and siting restrictions. Virginia Senate Bill 253, introduced by Senator L. Louise Lucas, would shift billions in grid upgrade costs from residential ratepayers to data centers using 25 MW or more, cutting average household bills by $5.52 per month while raising data center electricity rates by roughly 15.8%.

Maine's LD 307, which passed both House and Senate, would have imposed a moratorium on data centers above 20 megawatts until November 2027 and created a Maine Data Center Coordination Council to help municipalities evaluate projects, but was vetoed by Governor Janet Mills on April 24, 2026. State Representative Melanie Sachs, who sponsored the bill, said: "The beauty, I think, of this bill is to really take the lessons of what we've seen from the states around what happens with unfettered rapid development. They're now trying to claw back or change or retroactively put in place some policies. Maine has the opportunity to put that comprehensive framework in place to evaluate any trade-offs that we may want."

Representative Sachs also warned: "AI data centers are increasingly drawn to locations with available land and strong connectivity, qualities that Maine is well positioned to provide. But if these centers aren't thoughtfully planned and coordinated, they can place extraordinary demands on electric infrastructure, the surrounding environment and host communities." After the veto, she accused the governor of "resisting the will of a majority of Maine people."

Vermont's Senate bill S.205 proposes a moratorium on AI data centers above 100 megawatts through July 1, 2030, and remains before the Senate Finance Committee.

Seattle proposed an emergency 365-day moratorium on siting new data centers paired with a resolution for impact studies on city infrastructure, water usage, utility rates, land use, jobs, and public health, taking effect immediately upon adoption. Seattle City Council members said: "The bill will be paired with a resolution calling for impact studies of data centers on city infrastructure, water usage, utility rates, land use, jobs and public health." San Marcos, Texas, became the first Texas city to vote to block new data center developments in 2026, testing state limits on local authority despite state leadership moving to prevent counties from halting projects.

The NAACP released the Frontline Framework in 2026, the nation's first environmental and climate justice-focused data center principles, alongside a 2026 Playbook with recommendations for protecting frontline communities from AI data center harms. NAACP President Derrick Johnson declared: "These Recommendations are about accountability and protection. For too long, communities have been forced to absorb the environmental and economic costs of technological growth without safeguards or consent." Johnson added: "Innovation must not come at the expense of clean air, clean water, or affordable utilities. Our communities deserve to lead decision-making where they live in the AI era."

NAACP's Abre' Conner said: "Data centers are not abstract—and the patterns are that they are physical facilities that consume massive resources and too often pollute the same neighborhoods already burdened by industry." The NAACP sued Elon Musk's artificial intelligence company xAI in April 2026 over its use of methane gas turbines to power a data center in a suburb of Memphis, Tennessee, alleging unpermitted emissions in historically Black neighborhoods.

The Trump administration's executive orders on AI infrastructure do not override state authority on land use, zoning, or utility regulations, leaving states with material control over AI data center deployment. But states are addressing water consumption and energy infrastructure costs for facilities as small as 10 MW, creating a gap with federal rules that only apply to centers above 100 MW.

This patchwork of state and local action reveals the accountability gap: communities are improvising protections because the federal government has framed AI as national security rather than public infrastructure requiring democratic input.

The Window Is Closing

These three intertwined decisions—model access, chip export controls, and infrastructure regulation—are being made right now, in June 2026, by federal and corporate actors without public process. Once data centers are sited, grid infrastructure built, chip supply chains locked in by export rules, and model access restricted to national-security partners, reversing those decisions becomes nearly impossible.

Apple used WWDC 2026 to reposition its platforms as AI-centric operating systems, with Apple Intelligence becoming a systemwide layer across iOS 27, macOS 27, iPadOS 27, watchOS 27, and visionOS 27, scheduled for fall 2026 release. The embedding of AI into operating systems and the physical buildout of gigawatt-scale data centers represent long-term commitments that will shape regional economies, energy grids, and technology access for a generation.

States and cities are racing to regulate data centers before they arrive, but they lack coordination, federal support, or authority over chip exports and model access—leaving a fractured, reactive approach.

The core civic question remains unanswered: if AI infrastructure is essential to the economy—like highways, telecom, or electricity—shouldn't communities have a democratic voice in how it's governed, rather than leaving it to closed-door deals between Washington and Silicon Valley?

That conversation isn't happening at the federal level, and the window for forcing it open is closing as infrastructure gets built, contracts get signed, and the national-security frame hardens into permanent policy.