The expansion of AI infrastructure is encountering a widening set of constraints. Electricity supply, transmission capacity, water availability, permitting, tax incentives, and community acceptance are no longer secondary considerations; in many markets, they are becoming conditions for whether a project can proceed at all. The evidence is concentrated in July 2026, with the broader opposition trend extending back to April 73.
For Apple, the direct implication is more limited than for hyperscalers and dedicated data-center operators. One claim specifically states that Apple is not building data centers 73. Yet the issue remains strategically relevant. Apple’s access to cloud capacity and AI compute, the pricing of utility services, environmental regulation, and the cost of competing AI offerings increasingly depend on infrastructure controlled by hyperscalers and their partners. We must therefore distinguish between Apple’s direct construction exposure and its indirect dependence on an infrastructure system that is becoming more costly and politically contested.
The Infrastructure Constraint
Electricity and water as binding inputs
The most consistent conclusion is that data centers are unusually intensive users of both electricity and water, and that this intensity is becoming politically salient. Data centers consume substantial quantities of electricity and water 18,64,73, including clean drinking water in some operating models 1. Their global energy consumption has been estimated at 2% 49, while electricity use has reportedly surged 267% 78. Three Wisconsin developments alone are said to account for 72% of the state’s peak electricity load 44.
These demands are imposed on systems that must also serve households and conventional businesses. Back-to-back heat waves and continuous data-center demand are straining the grids on which both groups depend 19. Aging networks are approaching a tipping point 19. A particularly revealing operational event occurred in July 2024, when approximately 1,500 MW of data-center load disappeared almost simultaneously 55. The episode illustrates not only the scale of computational demand, but also the reliability and system-management risks associated with concentrating such demand in a small number of facilities or regions.
Water presents a parallel constraint. Cooling is a central operating dependency 7, and access to water can affect both project feasibility and financing 7. Water-related permitting is already embedded in the development process 7. In Virginia, a groundwater study of the eastern region 46,63 concluded that an evaporatively cooled data center could not reliably secure sufficient groundwater under current conditions 63. The study called for tighter water regulation 46 and emphasized that comprehensive public data on groundwater withdrawals remain unavailable 63.
The scale of Virginia’s industry makes this more than a localized concern. Virginia is described as the world’s largest data-center hub 63, with 371 facilities operating and another 438 planned 63. Most capacity lies within public-water service areas that draw primarily on surface water 63. The combination of extensive planned development and uncertainty surrounding projects in the eastern region 63 suggests that water availability could become a gating factor rather than a secondary environmental, social, and governance consideration. A Georgia facility reportedly used 29 million gallons of water 78, while state Senator Richard Stuart linked a local water shortage to new data centers 63.
Power availability, not demand alone
The relevant distinction is between the desire to build capacity and the ability to connect it to a functioning grid. Power availability—not merely projected demand—is becoming the binding constraint on data-center scaling economics 64. Large interconnections reportedly require five to seven years 69, and accommodating additional generation and data-center capacity will require major transmission expansion 3. Analysts increasingly describe data-center limits as a constraint on the IT industry itself 12, while the grid cannot support all currently planned facilities 73.
This creates a tension between the bullish outlook for physical data-center, power, and building activity 76 and the practical reality that grid-connection speed is now central to both compute expansion and the success of the energy transition 55,56. The constraint is not uniformly negative. Grid modernization and AI demand are supporting base-metal demand 61, and new opportunities are emerging in electricity supply and data-center solutions 15. But these benefits do not remove the underlying allocation problem. They indicate that the cost of adding capacity is moving upstream, into generation, transmission, cooling, and reliability infrastructure.
The Shift from Voluntary Assurances to Cost Allocation
Making large loads pay
Policy is evolving from voluntary assurances toward more explicit cost allocation and project restrictions. The White House convened data-center firms and utilities around a voluntary ratepayer pledge 60. The pledge was later described as having 187 signatories representing roughly 80% of U.S. generation, along with 23 governors 57. A separate initiative was also voluntary 58, while President-elect Trump was reported to have said that 200 utility companies had pledged to cap residential energy costs 65.
These commitments coexist with more prescriptive measures. Florida legislation requires large data centers to pay their own energy and water costs 58. A proposed federal framework would require private sourcing of electricity and water rather than reliance on public utilities 58. Byron Donalds proposed legislation intended to prevent AI facilities from increasing utility costs 25. Donalds said builders were willing to absorb utility costs and emphasized efficient construction while protecting Florida’s ratepayers and water resources 58. The political compromise is therefore increasingly framed as “build, but pay.”
The clearest regulatory direction is the transfer of incremental infrastructure costs from households to the large loads that create them. FERC opened a “just and reasonable” review of interconnection-cost allocation across six grid operators 57. PJM stakeholders advanced a plan requiring data centers to pay directly for new generation 55, while Virginia regulators considered requiring data centers to cover transmission costs 55.
The proposed and enacted Ratepayer Protection Act framework would amend PURPA for facilities drawing at least 100 MW 57. It would require those facilities to fund transmission and generation upgrades 57, shifting the burden from ratepayers to data-center operators 57. Six states reportedly already have laws requiring data centers to pay their share of grid costs 60. Portland General Electric offers an early market example: data-center rates rose roughly 29% while residential bills fell by as much as 2.1% 55. Federal regulators have also directed NERC to develop mandatory reliability standards for large data centers and other computational loads 55.
The economic significance is straightforward. Where a large load once benefited from socialized network investment, the marginal cost of generation, transmission, and reliability is increasingly being assigned to the facility itself. This does not necessarily prevent construction. It changes the equilibrium by raising the capital required, lengthening negotiations, and favoring operators able to secure power and finance infrastructure over longer time horizons.
Subsidies, disclosure, and the evidentiary standard
The same pressure is visible in the treatment of subsidies and disclosure. Several states are reviewing data-center tax incentives because of grid strain 64. New York opened a review of sales-tax exemptions 60 and imposed a one-year pause on new data-center tax exemptions 60. Critics argue that tax breaks have starved schools 27, while states are beginning to count more fully the costs of water use, grid investment, and foregone public revenue 27.
Pennsylvania now requires Big Tech to disclose data-center energy and water usage 31. Maryland’s Office of People’s Counsel opposed an emergency hosting-capacity reporting waiver for Potomac Edison without a sufficient review of stakeholder impacts 9. The direction of travel is thus toward greater transparency, direct cost recovery, and evidence-based permitting. Public data remain incomplete, however 63. That limitation matters: sound allocation requires not only a willingness to charge for marginal infrastructure, but also reliable measurement of the resources a project consumes and the costs it imposes.
Permitting and the Politics of Siting
New York’s moratorium
New York represents the strongest state-level intervention identified in the evidence. Governor Kathy Hochul signed Executive Order 62 on July 14, establishing a one-year moratorium on new hyperscale data centers requiring at least 50 MW 20,45,47,50. The order directs the Department of Environmental Conservation not to issue discretionary environmental permits 45 and blocks permits for facilities above 50 MW for up to one year 60. New York has been described as the first U.S. state to implement a hyperscale moratorium 47.
This is not an isolated administrative decision. It forms part of a wider state and local trend 20. Oklahoma reported four projects cancelled or placed under moratorium 75. Florida municipalities imposed bans of at least one year 58, and more than 20 counties or municipalities rejected major facilities 58. Mississauga approved a one-year moratorium on AI data centers 17.
Local zoning is becoming another instrument of adjustment. Measures in Summit, New Jersey, would remove data centers as a principal use in certain industrial zones 4, cap accessory data-center activity at 12,000 square feet 4, and pursue changes intended to protect local employment and community character 4. These measures do not amount to a uniform national prohibition. They do, however, make the geography of compute more dependent on local institutional conditions.
Speed versus legitimacy
The investment consequence is not simply slower permitting. It is a higher and less predictable cost of bringing capacity online. No entity has completed a gigawatt-scale data center within one year 74, while many facilities reportedly remain unbuilt with hardware sitting on shelves 77. Regulatory acceleration zones may bypass land permitting and community consultation 41, but business groups and other stakeholders oppose mandatory restrictions 26. They warn that regulation could stifle competition, drive innovation overseas 62, and produce international investment losses 26. A counterweight is the $15 million reportedly provided by workers and grassroots groups opposing regulation 59.
The underlying conflict is between speed and legitimacy. Faster construction may improve compute availability in the short run, but bypassing local consultation can increase opposition and litigation in the longer run. Conversely, a more deliberate process may raise initial friction while producing projects with greater community acceptance. The eventual industrial structure will depend on which of these adjustment paths proves less costly.
Community Opposition and Distributional Concerns
Opposition is broadening beyond utilities and environmental organizations. Seven in ten people reportedly oppose new data centers 2, and opposition in the United States is growing 35,72,73. Residents and local leaders cite utility bills, disruption, land use, water consumption, carbon footprints, and environmental damage 53,58. Noise complaints are also emerging 40, alongside disputes over housing displacement, including the risk to Myron Manor associated with a proposed Mt. Pleasant development 36. A proposed regulation would prevent data centers from displacing housing 26.
Community participation is visible in large hearing sign-ups and City Hall processes 36,43. A petition addressing privacy, labor markets, climate, and local communities exceeded 20,000 signatures 30. Claims that Big Tech has hidden community impacts 31, and that no technology executives live within five miles of a data center 34, reinforce the perception that benefits and burdens are distributed unevenly. These latter claims are single-source or advocacy-oriented rather than consensus evidence, but they are nevertheless relevant to the political economy of siting. Perceived distance between decision-makers and affected communities can increase the friction attached to each additional project.
Corporate disputes provide concrete examples of this friction. Meta is contesting a Wyoming town over a water case 39, including a $10,000 fine 39, and argues that it should not bear cleanup costs 39. Its Louisiana project reportedly expanded from $10 billion to $50 billion in less than two years 38, while critics say Meta benefits from tax write-offs 38. Meta separately claims to cover its own power costs 38.
Meta and BlackRock announced a $14 billion El Paso project 51, with approximately 1 GW of capacity 13 and more than $12 billion of debt led by BlackRock 13. These figures demonstrate both the scale of the infrastructure opportunity and the financial exposure to cost overruns, leverage, permitting delays, and resource constraints. Other projects have attracted praise from state and business leaders and included community investments 37, suggesting that local benefit-sharing can mitigate, though not eliminate, opposition. A Michigan campus was described as a $16 billion investment 37, and data-center construction is contributing to an accelerating business-investment cycle and broader U.S. industrial-capacity expansion 22,67.
Political and Environmental Counterforces
The coalition opposing unchecked expansion is unusually cross-partisan. Progressive climate activists and conservative populists have both focused on data centers 57,58. Local criticism is centered on household bills, disruption, and environmental damage 58. Florida is particularly instructive because resistance has developed alongside Republican proposals that support construction while shifting costs to operators.
The broader political environment adds further complexity. Protests have targeted Google’s military and ICE contracts 32, while concerns have been raised about covert industrial activity and technology theft 14. The United States has imposed restrictions on Chinese technology 21 and pursued efforts to contain China’s technology ambitions 16. Some claims in the cluster—protests against Trump’s global trade transformation 29, attacks associated with “woke capitalism” and red-state ESG boycotts 6, and opposition by 12 states to a separate merger 23,54—are peripheral to Apple and to data-center economics. They nevertheless reinforce the broader point that technology infrastructure is being drawn into national-security, cultural, and industrial-policy disputes.
Environmental and efficiency standards are also moving toward renewable power. A United Nations dialogue proposed that all data centers be powered by renewable energy by 2030 5, while proposed industry standards set a similar target 5. Advocacy groups warn that electricity bills could rise rapidly unless Big Tech uses 100% new renewable energy 37,42. The industry says that environmental stewardship and corporate responsibility are increasingly embedded in design and procurement 50. Advanced UPS deployment is being targeted across multiple U.S. hyperscale campuses 48.
Renewable procurement and storage may improve project acceptability and reduce exposure to carbon regulation and grid volatility. They may also raise near-term capital costs. More importantly, they do not automatically solve transmission bottlenecks or water scarcity. We must distinguish between a project’s source of electricity and its ability to obtain a reliable physical connection, just as we must distinguish between reducing carbon intensity and securing sufficient cooling water.
Evidence That Requires Caution
Several claims are isolated and should not carry the same weight as the more consistently supported evidence on grids, water, and regulation. The description of data centers as currently largely military-related 71, the claim that a riot could result from a nearby facility 70, allegations of contaminated water harming children 33, and the assertion that space-based data centers proposed by Jeff Bezos and Elon Musk would be catastrophic 35 are not corroborated at the same level.
The cluster also includes adjacent technology-governance issues: criticism involving SCHUFA 24, transatlantic data-transfer exposure and compelled access to European servers 52, privacy advocacy by NOYB 11, and a U.S. ruling affecting transatlantic data transfer 8. These issues may matter to technology companies, but they are not direct evidence about the economics of data-center expansion. Similarly, the isolated claim that thousands of data-center controllers are open to takeover 28 and the claim that ASML faces cloud resistance 10 should not be treated as central investment conclusions without additional corroboration.
Implications for Apple
Direct and indirect exposure
Apple’s direct infrastructure exposure appears lower than that of Meta, Microsoft, Google, and dedicated data-center operators, given the specific claim that Apple is not building data centers 73. This distinction may be strategically valuable. Apple may avoid some of the headline risk associated with hyperscale construction, local water disputes, tax subsidies, and direct allocation of grid costs.
The indirect exposure is more substantial. Apple relies on cloud and compute suppliers, has its own AI ambitions, and competes in markets whose economics are shaped by the cost and availability of third-party infrastructure. If power, water, or permitting becomes scarce, suppliers may prioritize their largest and most financially committed customers. Apple could consequently face higher AI-service costs or slower deployment, even without bearing the permitting and resource liabilities of a hyperscale developer.
The value of efficiency and on-device processing
The cluster points to a two-sided implication for Apple. Defensively, Apple’s comparatively limited data-center construction footprint may reduce regulatory and community liabilities relative to hyperscale peers. Offensively, constrained compute could increase the economic value of efficient models, edge processing, and hardware-software integration.
This is where the distinction between centralized and distributed computation becomes important. The infrastructure bottleneck strengthens the relative appeal of a device-centered, privacy-oriented AI strategy if such a strategy requires less centralized inference per unit of functionality. The claim that frontier models bear massive data-center costs 68 supports this interpretation, although it is a single-source observation. The conclusion is therefore conditional: infrastructure scarcity does not establish that on-device AI will prevail, but it raises the marginal value of architectures that reduce dependence on scarce centralized capacity.
What investors should monitor
For Apple investors, the central variables are not simply whether data-center demand continues. The more useful questions are who pays for incremental power, how quickly capacity can be approved, and which customers receive access when supply is constrained. Direct cost-shifting under the Ratepayer Protection Act framework 57, moratoria for facilities above 50 MW 47, renewable-power requirements 5, and disclosure obligations 31 could raise the cost of AI infrastructure across the industry.
At the same time, large projects such as Meta’s El Paso development 51 and the continuing Virginia pipeline 63 show that capital remains available and that the buildout has not stopped. The appropriate conclusion is therefore a moderation of AI-infrastructure growth assumptions rather than a collapse thesis. Constraints are more likely to lengthen timelines, raise costs, and favor operators with secure power, water, financing, regulatory access, and credible community-benefit agreements.
Competitive divergence across the ecosystem
The regulatory backlash may produce divergent outcomes across the industrial ecosystem. Dedicated data-center owners and infrastructure suppliers may benefit from durable demand and grid-modernization spending, as reflected in institutional interest in Iron Mountain ahead of earnings 66. They also face approval and resource risks. Caterpillar is specifically described as vulnerable to local restrictions through higher costs, slower investment, additional approval hurdles, and reduced site availability 20.
Apple is not directly identified as either a beneficiary or a casualty of those restrictions. Its relative advantage, if it materializes, would instead come from requiring less centralized infrastructure per unit of AI functionality and from shifting workloads toward devices or more efficient models. That remains an analytical inference rather than a directly reported fact.
Conclusion
Data-center expansion is evolving from a straightforward growth narrative into an infrastructure-allocation problem. Electricity, transmission, water, permitting, and community consent are becoming binding constraints 63,64,73. Regulation is increasingly designed to make hyperscale operators pay for incremental grid and resource costs, with New York’s 50-MW moratorium and the 100-MW Ratepayer Protection Act framework as prominent examples 47,57.
Apple’s direct exposure appears lower because it is not identified as building data centers 73. Its indirect exposure remains material because its AI ambitions and competitive position depend on compute supplied by an increasingly constrained infrastructure base. The likely consequence is not the disappearance of AI investment, but a slower and more discriminating process of capacity formation.
Under current conditions, the evidence supports a measured investment stance: moderate assumptions about the speed and cost of AI scaling, and place greater weight on companies able to secure power, water, financing, regulatory access, and community consent. For Apple, the most important strategic question is whether its hardware-software integration and potential reliance on efficient, on-device processing can convert infrastructure friction from a cost imposed by the ecosystem into a relative competitive advantage.