The Next Bottleneck for U.S. AI Compute: Not Chips or Power, but Permits

Barclays' latest research report points out that as data centers across the United States face increasingly strong community opposition, permitting delays, and regulatory fragmentation, "permitting" is becoming the next core bottleneck for AI infrastructure expansion, with an impact potentially comparable to compute shortages or insufficient power supply.

A Gallup poll in March this year showed that 71% of U.S. respondents opposed building data centers near their communities, a proportion even higher than the 53% who opposed nuclear power plants. The opposition spans party lines, with majorities of Republicans, independents, and Democrats all opposed, and according to a more recent Politico survey, this sentiment is continuing to worsen. Barclays believes that today's political and permitting decisions will directly affect AI compute supply in the latter half of this decade. If permitting challenges and community resistance continue to slow the deployment of hyperscale data centers, scarce compute resources will become even more valuable.

Although the Trump administration has positioned AI infrastructure as a strategic economic and national security asset and is actively advancing it through multiple policy tools such as opening federal lands, expanding transmission, reforming electricity markets, and deploying nuclear energy, many of the most critical decision-making powers remain dispersed at the local level, with federal influence constrained by the highly fragmented U.S. utility, permitting, and regulatory systems.

Opposition Is Broad and Bipartisan

The politicization of data centers is becoming a major risk to the AI buildout wave. The Barclays report, citing Gallup survey data, notes that opponents' concerns are not focused on a single issue but span electricity consumption (18%), water usage (18%), quality-of-life impacts (22%), rising electricity bills (20%), environmental pollution (16%), employment impacts (14%), and concerns about AI technology itself (14%).

This broad opposition reflects a deeper social perception: local communities bear the costs while the benefits flow elsewhere. In some projects, confidentiality agreements and limited public disclosure have further exacerbated community distrust regarding transparency. At the same time, overall public skepticism about AI technology, privacy security, and the social influence of big tech companies is reinforcing this resistance.

Political sensitivity has already triggered substantive policy responses. The National Republican Senatorial Committee recently warned AI companies that public opposition to data centers could jeopardize a key Senate seat in Ohio. Texas Governor Greg Abbott has asked state utility regulators and grid operators to review data center projects before advancing interconnection procedures; Pennsylvania Governor Josh Shapiro issued an executive order requiring developers to meet state infrastructure standards and obtain local approval before permit applications are reviewed; New York Governor Kathy Hochul issued an executive order in July 2026 establishing the nation's first statewide moratorium on new hyperscale data center projects; and in Virginia—the world's largest data center market—Governor Abigail Spanberger has required data centers to pay for transmission infrastructure dedicated to serving their facilities rather than shifting costs to other electricity users.

Rising Electricity Bills Become the Most Politically Potent Issue

Among all the controversies, electricity affordability has become the most politically sensitive core issue. In the first quarter of 2026, the average U.S. residential electricity price rose 12% compared with the same period in 2024, with even more pronounced increases in regions experiencing rapid data center expansion—Washington state up 24%, Virginia and Pennsylvania both up 17%, and Ohio, Louisiana, and Illinois all up 14%.

Barclays notes that these increases cannot be entirely attributed to hyperscale data center expansion; a considerable portion reflects years of underinvestment in grid infrastructure and grid-hardening expenditures to cope with extreme weather. However, the rapid growth in data center demand is indeed putting greater pressure on utilities and regulators to balance reliability, affordability, and investment incentives.

The report is also skeptical of the argument that "data centers can lower electricity prices by expanding the customer base and spreading fixed costs." This logic holds only if there is ample redundancy in existing generation and transmission capacity, which is not the case in most regions where AI data centers are located. Data center cooling demand tends to peak in summer, overlapping heavily with the periods when the grid is under the greatest stress from air-conditioning loads. Meeting incremental demand therefore typically requires building new dispatchable generation and transmission upgrades rather than tapping existing idle capacity.

Regulatory Landscape Fragmented, Local and State-Level Tensions Intensify

Barclays categorizes data center governance models across U.S. states into six types, revealing an increasingly complex regulatory map.

Under the state-led preemption model, West Virginia passed the Electricity Generation and Consumption Act in 2025, restricting local government regulatory authority over qualifying high-impact data centers, moving primary approval authority to the state level, and pairing it with tax incentives and a microgrid certification program.

Under the direct state regulation model, New York's statewide moratorium is currently the most typical case, but a similar moratorium passed by the Maine legislature was vetoed by Governor Janet Mills, and similar proposals in Minnesota, New Hampshire, Oklahoma, and South Dakota all failed to advance. Under the hybrid state-local regulation model, local governments in Texas, Ohio, and Pennsylvania retain primary zoning authority, but state agencies are increasingly influencing project outcomes through grid planning, interconnection requirements, and customer protection policies.

This fragmented landscape means permitting risk is becoming highly localized, and regulatory geography has become an important variable in project siting, development timelines, and the pace of AI infrastructure deployment.

Federal Influence Limited, Key Decisions Remain Local

The Trump administration continues to frame AI infrastructure as a strategic economic and national security priority. The White House's AI Action Plan released in July 2025 lists AI leadership as a dual economic and national security priority, with a dedicated pillar to accelerate AI infrastructure deployment. Trump himself stated on Truth Social that he "never wants Americans to pay higher electricity bills because of data centers" and pushed hyperscale operators, utilities, and state officials to sign the Customer Protection Pledge, requiring data center developers to build, bring, or buy their own power and bear all associated infrastructure costs. However, the pledge is voluntary, non-binding, and does not address core issues driving local opposition such as water use, land use, emissions, noise, and community impacts.

On federal land use, the U.S. Department of Energy has identified multiple potential AI infrastructure sites, including the PORTS-Pike Technology Park in Ohio with up to 8 gigawatts of compute capacity (jointly involving SoftBank, OpenAI, Nvidia, and AEP Ohio), the former Paducah gaseous diffusion plant site in Kentucky with over 1.2 gigawatts (led by Brookfield and NextEra Energy), and a 1-gigawatt project at the Savannah River Site in South Carolina (undertaken by Amentum).

However, federal influence is fundamentally constrained by the highly decentralized U.S. utility and regulatory system. The most critical decisions—transmission investment, interconnection rules, cost allocation, reliability standards, and customer protections—remain dominated by regional electricity markets and state-level regulatory processes. While FERC has advanced measures such as Order 1920 (long-term regional transmission planning) and Order 2023 (interconnection reform), building a truly national transmission network still requires resolving complex interest coordination across multiple states, multiple utility territories, and multiple regulatory jurisdictions.

Can Hyperscalers' Community PR Offensive Succeed?

Facing mounting community resistance, hyperscale operators such as Microsoft, Amazon, Google, Meta, and OpenAI are increasing community engagement investments, attempting to trade economic benefits for social license. Microsoft launched a "Community-First AI Infrastructure" program, pledging not to increase existing customers' electricity bills, minimize water consumption and replenish excess water use, create local jobs, expand local tax revenue, and invest in AI education and digital skills training. Meta's Hyperion campus in Louisiana is accompanied by large-scale power infrastructure investment, expected to save Entergy Louisiana customers approximately $2.7 billion in electricity costs; the PORTS-Pike campus sponsors have committed approximately $4.2 billion for grid and transmission upgrades.

However, Barclays believes that many of the issues driving local opposition are structural rather than reputational. Even if hyperscalers make credible grid upgrade and emissions reduction commitments, electricity prices in many parts of the U.S. may continue to rise for reasons unrelated to their actions. Water disputes involve multidimensional trade-offs among direct water use, indirect upstream water use, and emissions from the power sector, far more complex than they appear on the surface. In the current AI arms race environment, the priority of "rapid access to power" has overtaken cost and emissions control, making it difficult for the industry to demonstrate substantive progress on long-term sustainability goals to the outside world.

Bring-Your-Own-Power and Emerging Solutions: Mitigation, Not a Cure

Facing permitting and grid bottlenecks, Bring-Your-Own-Power (BYOP) and Power-as-a-Service (PaaS) models are gaining increasing attention. By building dedicated generation facilities, developers can reduce reliance on transmission upgrades and lengthy interconnection queues while lowering the risk of shifting grid upgrade costs to other electricity users. In the AI compute race, the value of deployment speed has surpassed the cost of power itself—a single gigawatt of AI compute can support tens of billions of dollars in annual revenue, making deployment delays extremely costly.

However, Barclays emphasizes that BYOP is not a panacea. Hyperscalers generally still prefer to connect to the grid when conditions allow, because the grid's economies of scale and diversified supply mix are more conducive to achieving the "five nines" (99.999%) reliability required for mission-critical loads. Most BYOP solutions rely heavily on natural gas generation and associated pipeline infrastructure, making it economically difficult to achieve comparable reliability levels. Moreover, BYOP projects face the same risks of air permits, pipeline capacity, fuel supply agreements, and community acceptance—the controversy surrounding xAI's Memphis facility is a prime example.

From a longer-term perspective, compute efficiency improvements, advanced energy technologies (including small modular nuclear reactors), distributed computing architectures, and even space-based data centers are seen as potential paths to alleviate ground-based infrastructure constraints. But Barclays notes that most advanced nuclear technologies remain years away from commercial-scale deployment and are unlikely to relieve the industry's near-term power and interconnection pressures. Data centers can relocate, but bottlenecks often follow—perhaps the most apt footnote to the current AI infrastructure expansion dilemma.