Meta’s AI strategy is now constrained by infrastructure, not ambition. The company is committing tens of billions of dollars to data centers across Louisiana, Texas, Alberta, and other regions. The core risk is straightforward: Meta must secure reliable power, water, permits, financing, and community consent before those facilities can generate returns.
The scale is significant. Meta’s Richland Parish, Louisiana, project is described as a $50 billion investment and one of the largest single-site investments in U.S. history 31,32. Its El Paso campus is estimated at approximately $14 billion and up to 1,000 megawatts of demand 4,34. Meta is also developing an Alberta facility valued at roughly $13 billion, reportedly the largest data-center project in Canadian history, with approximately 3,300 expected jobs 1,8,16.
This buildout could give Meta proprietary control over compute capacity, storage, and infrastructure design. It could also support future cloud or compute-marketplace offerings. But control is not free. Energy procurement, water stewardship, permitting, local ratepayer exposure, and financing structure are becoming direct drivers of execution and valuation.
Meta is prioritizing capacity and data-center shells in 2026 and 2027 while reportedly deferring substantial server purchases until 2028 5,45,71. That approach preserves optionality. It also creates a timing risk: fixed infrastructure commitments may arrive before demand and monetization are fully established.
The Infrastructure Constraint
Power demand is reaching utility scale
Meta’s largest facilities are no longer ordinary commercial loads. The Louisiana campus is projected to require approximately 5,000 megawatts, roughly comparable with New York City’s average electricity load. It could require about 7.5 gigawatts of new generation capacity and ten gas-fired plants, each rated at 754 megawatts 33,51.
Entergy’s April 2026 application includes new transmission infrastructure and battery storage. The project is structured around a 20-year electricity-service contract 33. The weakness is in the assumptions. The application describes the assumed load factor as very high for any load, while the actual expected load factor remains confidential 33. Entergy reportedly did not test lower-demand scenarios or meaningfully evaluate flexible load management after Meta indicated that it was not interested 33.
The conflict is structural. Meta wants firm, high-utilization power for AI workloads. Utilities and regulators have an incentive to preserve flexibility and limit stranded-asset risk. If demand falls short, ratepayers and utilities may be left carrying infrastructure built for Meta’s forecast rather than its actual consumption.
El Paso presents the same issue at a smaller but more concentrated scale. The planned campus would require approximately 1 gigawatt 36,50,60, equivalent to roughly 40% of El Paso Electric’s electricity consumption from all other customers during a hot summer afternoon 13. The proposed $500 million McCloud plant would provide 225 megawatts of continuous power 13,34. It would serve only Meta during its first five years and remain disconnected from the broader grid used by El Paso residents 34.
That arrangement isolates some reliability and ratepayer exposure. It does not eliminate the broader system requirement. El Paso Electric expects to construct additional generation over time to meet the campus’s full demand 34. Dedicated generation is therefore a partial hedge, not a complete solution.
Meta’s power strategy is diversified. Reported arrangements include dedicated natural-gas supply for Alberta 2,17, nuclear power-purchase agreements totaling more than 7.7 gigawatts 67, and a broader mix of nuclear, gas, and renewables 62. One proposed power architecture also includes solar generation, battery energy storage, and UPS protection, with a stated 99.999% uptime objective 52. These measures can improve resilience and reduce reliance on diesel backup systems 52.
They do not make the system automatically low-carbon. Gas generation increases emissions and may trigger carbon-capture or EPA compliance requirements 35. Nuclear arrangements carry safety, waste, licensing, and public-acceptance risks 11,25. The math is simple: reliability, emissions, and cost must be evaluated together.
Ratepayer exposure is the local pressure point
The Louisiana project could increase residential electricity bills by approximately $8–$13 per month, according to a commission consultant 51. In El Paso, the facility is expected to absorb at least $40 million annually in fixed-system costs for El Paso Electric 34. These are single-source estimates, not broad consensus. They align with wider claims that large data-center projects can raise local electricity costs and shift system costs to households 10,12,13.
The proposed El Paso special rate and credit guarantee are designed to shift project-specific risks away from existing customers 34. The associated power infrastructure remains subject to Public Utility Commission review 13. Texas regulators retain oversight authority, and binding requirements could require Meta to comply 34.
The financial structure reduces Meta’s upfront capital burden but increases complexity. The El Paso venture with BlackRock is described as an asset-light arrangement in which institutional investors and private-credit providers assume more capital-expenditure obligations 60. Meta contributed land and partially completed assets valued at $2.3 billion 13. The facility is reportedly leased back to Meta for 20 years 36.
A similar structure for Hyperion transferred 80% ownership to Blue Owl Capital while Meta retained 20% 13. Meta had approximately $279 billion of uncommenced data-center lease commitments as of August 11, 2026 63. One facility is financed with $12.3 billion of project bonds, exposing the structure to debt-market conditions and investor appetite 13.
Leasebacks can improve near-term capital efficiency and return on invested capital. They also create long-duration fixed obligations and future operating expenses. The unresolved issues are control rights, beneficial ownership, consolidation, and structured-finance disclosure 14,40. Asset-light financing does not remove economic exposure. It repackages it.
Tax incentives sharpen the distributional debate. The El Paso transaction includes an 80% property-tax break 13. The Richland Parish project benefits from sales-and-use tax exemptions that may last 20 years, with a possible ten-year renewal 51. Meta is estimated to receive $3.3 billion in data-center-related tax exemptions 51.
These incentives can support development, construction activity, tax revenue, and skilled employment. Meta’s projects could create thousands of construction and operating jobs, including approximately 3,300 jobs cited for Alberta 8,13. But the benefits remain contested when tax abatements coexist with higher utility bills and public infrastructure costs. That imbalance is the core issue identified in El Paso 13.
ESG Risk Is Measured Locally, Not in Corporate Averages
Water and cooling claims require facility-level scrutiny
Data-center electricity and water consumption are now material ESG and operational issues 9,26,27,28,61. Meta has set a goal of becoming water-positive by 2030 and replenishing 200% of water consumed in high-water-stress areas 3,6.
Its Central Texas AI facility uses closed-loop liquid cooling that reportedly requires only a one-time fill 39. Meta says its water consumption will be below that of a typical golf course in Temple, Texas 39. These are meaningful mitigants. They do not close the argument. Public commenters have questioned whether closed-loop systems fully address heat impacts, while community members continue to raise concerns over water and energy use 39. Water remains central to Meta’s internally built data-center model and to the engineering challenge of cooling high-density AI clusters 42.
The broader accounting problem is more important than the headline metric. Sustainability outcomes depend on workload timing, grid carbon intensity, procurement structure, facility design, cooling technology, local water conditions, and reporting methodology 68. Annual renewable-energy matching can diverge materially from physical, interval-level consumption 68. Low-cost electricity may reflect a fossil-heavy grid rather than superior sustainability 68.
Scope 2 emissions can differ by nearly 100 times between high- and low-carbon grids 38. Embodied and other Scope 3 emissions can exceed 40% of lifetime emissions in facilities powered by low-carbon electricity 38. Similar power-usage-effectiveness figures therefore do not guarantee similar environmental outcomes 68.
Investors should assess Meta’s environmental disclosures against facility-level, time-matched power and water data rather than corporate averages. Renewable procurement does not eliminate embodied emissions or indirect water use 38. On coal- or nuclear-heavy grids, indirect water consumption can exceed on-site use 38. A credible framework would combine grid carbon intensity, time-matched electricity use, PUE, water-consumption effectiveness, cooling requirements, resilience, procurement quality, and disclosure transparency 68. Continuous monitoring and auditable links between reported sustainability performance and actual grid conditions would strengthen credibility 68.
Community consent is becoming a permitting asset
Community opposition is no longer a secondary reputational problem. Residents are challenging hyperscale developments over electricity costs, water consumption, noise, and land use 59. Opposition can create permitting friction and additional stakeholder-engagement obligations 58. Residents near one Meta facility reported persistent noise and vibrations 37. Wider concerns include lighting, setbacks, traffic, emergency-response coordination, ownership, financing, and local quality of life 29,43,66.
Hyperscale campuses can occupy hundreds of acres, replace farmland or natural landscapes, fragment habitats, and increase stormwater-management requirements 64. Their national environmental footprint may be modest relative to agriculture, beef production, golf courses, or manufacturing 41. That comparison does not resolve local opposition. Averages do not pay local utility bills or restore local water supplies.
Municipalities are increasingly evaluating data centers against long-term electricity and water availability, environmental effects, and community impacts 66,69. Texas proposals would require more disclosure, including projected electricity demand 29. State guidelines are intended to prevent facilities from straining energy and water resources 23. Meta, Google, and OpenAI have committed to following those standards 23, and Meta has publicly endorsed the relevant Texas guidelines 18,19.
Meta is building a social-license strategy. It has proposed a $1 billion Future Is for Everyone Fund for host communities, focused on job creation, schools, and public services 21,53,70. It has also committed to local skilled-labor training 55. A proposed community-benefit policy would allocate 20% of newly generated power to local community use 65.
These commitments can reduce opposition and improve project durability. They do not substitute for transparent cost allocation, credible water and emissions reporting, or formal community-impact reviews. Backlash remains a tail risk for both data-center facilities and associated energy infrastructure 65.
Strategic Upside Versus Utilization Risk
Meta’s vertically integrated infrastructure strategy gives it control over compute capacity, storage, and control-plane design 42,57. Infrastructure scale is a stated component of the positive investment thesis 47. The buildout supports high-density GPU clusters and could eventually support cloud or compute-marketplace offerings.
Meta’s internal demand could provide a baseline for a proposed compute-auction market 54. Such a market would require continuous participation from thousands or tens of thousands of bidders 54. Meta has not provided detailed public disclosure on a potential future cloud business 56. Commercial success would also depend on developing a competitive large language model 46.
There is a countertrend toward decentralized inference. Meta’s Muse Glimmer model supports offline and local deployment, potentially reducing centralized cloud, networking, and data-center requirements 15,20,44. Local inference shifts power consumption toward a large installed base of devices and transfers hardware, maintenance, security, and update responsibilities to users or developers 7,30. The net environmental effect is uncertain.
Orbital, underwater, and remote-desert data centers remain exploratory responses to terrestrial constraints rather than established commercial substitutes 22,28,37. The practical contest remains terrestrial: who can secure permitted, reliable, and socially acceptable power first?
Global data-center electricity consumption reportedly doubled between 2017 and 2024 38. Power availability and intensity are therefore becoming macro-level determinants of technology investment 24. Suppliers of generation, storage, cooling, and energy-efficiency technologies stand to benefit. Caterpillar, for example, is seeing increased demand for power-generation equipment from AI data centers 48.
The same trend creates policy risk. Moratoria or stricter standards could redirect growth toward alternative states and power markets 49. For Meta, access to power may become as important as access to GPUs. Control is the prize, but only if the controlled asset can operate at an economic return.
Implications for Investors
Meta’s data-center program is an infrastructure platform, not merely a capital-spending cycle. The company is securing long-lived physical capacity ahead of fully visible AI demand through proprietary construction, dedicated generation, long-term utility contracts, power-purchase agreements, and asset-light financing. This can preserve strategic control while limiting near-term balance-sheet intensity. It also exposes Meta to fixed commitments, project completion risk, utility regulation, interest rates, and eventual capacity utilization.
The immediate execution risks are power availability, permitting, and construction. Alberta faces potential delays, cost overruns, political scrutiny, and constraints on power, land, and water 8. El Paso depends on completion of the McCloud plant and related infrastructure; regulatory denial or delay would represent a severe project risk 13,34. More generally, insufficient power or water could delay Meta’s infrastructure program 56.
Louisiana has a different risk profile. Its power plan appears substantial, but the lack of lower-load sensitivity analysis and reliance on fossil generation could increase regulatory and ratepayer opposition 10,33.
The investment conclusion is clear. Meta’s scale is a competitive asset, but the marginal economics of AI capacity now include transmission, generation, cooling, water, community benefits, taxes, and financing costs. A technically feasible project can still become uneconomic or delayed if regulators require ratepayer protections, closed-loop water systems, emissions controls, or additional community compensation.
Meta’s scale, long-term contracting, and willingness to fund generation and community programs may give it an advantage over smaller AI operators that cannot internalize these costs. But the central uncertainty remains demand. AI utilization must grow quickly enough to justify infrastructure secured today.
Investors should track four indicators:
- Final regulatory approval and cost recovery for dedicated power projects.
- Actual versus assumed load factors and the treatment of flexible demand.
- The progression of leaseback and project-finance obligations.
- Facility-level reporting on hourly electricity emissions, water consumption, cooling performance, and community benefits.
Meta’s decision to maximize shells and capacity in 2026–27 while deferring some server purchases to 2028 provides optionality 45. It also puts infrastructure commitments ahead of definitive evidence of long-term compute monetization.
Bottom Line
Meta is attempting to build a moat around AI compute by controlling the physical infrastructure beneath it. Louisiana may require 7.5 gigawatts of new generation and ten gas plants, while El Paso’s 1-gigawatt campus could equal 40% of the utility’s other hot-afternoon load 13,51. Those numbers define the risk.
Asset-light ownership and 20-year leasebacks can reduce upfront capital intensity, but they create substantial fixed obligations and demand greater transparency around consolidation, control, and ESG performance 36,40,60,63. Water-positive commitments, closed-loop cooling, nuclear procurement, and renewable power are useful mitigants. They do not eliminate grid-carbon, embodied-emissions, indirect-water, or community risks 3,6,38,39.
The key valuation variable is execution quality. Meta’s infrastructure scale is strategically valuable. Permitting, ratepayer protection, community consent, financing discipline, and future utilization will determine whether that scale compounds shareholder value or becomes costly excess capacity. The best hedge is ownership—but only of assets that regulators permit, communities accept, and customers use.