Consider the circuit. The question is no longer merely whether demand for cloud computing and artificial intelligence will grow, but whether electricity, transmission, water, land, skilled labor, financing, and political permission can scale with it. Data centers currently account for approximately 1–2% of global electricity consumption 2,3,45. The Electric Power Research Institute estimates that they could reach as much as 9% of U.S. electricity consumption by 2030 4,70, while BloombergNEF projects U.S. data-center power demand to rise from approximately 35 GW in 2024 to 78 GW in 2035 1,37.
The range of forecasts is itself a material finding. Other BloombergNEF estimates imply that data centers could consume approximately 20% of U.S. electricity supply by 2035 37 or require 194 GW of power 37—figures materially more aggressive than the 78 GW projection 1,37. The evidence does not establish which estimate is most dependable. One source explicitly concludes that no single reliable forecast exists for the pace or magnitude of future data-center electricity demand 59. For Alphabet, infrastructure availability and capital discipline are therefore becoming as important as model performance and cloud demand.
Electricity and Grid Capacity
Power is becoming a strategic input
The investment race for AI models and data centers is intensifying competition for electricity 6,41. A hyperscale facility is not simply a large building containing servers. It is a concentrated, high-load electrical system requiring generation, substations, transformers, switchgear, transmission connections, distribution assets, backup systems, and specialized electrical construction 68. In some cases, power infrastructure may grow faster than the revenue associated with compute integration 68. The consequence is a structural demand shock to the U.S. power system 59, with data centers already a major contributor to higher electricity-demand forecasts 59.
The Department of Energy’s draft 2026 transmission study, synthesizing more than 120 studies 32, indicates that transmission planning is shifting from maintaining reliability and integrating new supply toward accommodating major new demand centers 32. AI data centers are described as fundamentally changing U.S. transmission investment 32, with rapid load growth led by AI, manufacturing, and other industrial users 32. They treat the grid as a simple bus, yet every interconnection is a resonant cavity with its own impedance, reactance, and transient response.
Regional evidence makes the constraint concrete. PJM forecasts 70 GW of large-load growth by 2038 61, even as generation retirements collide with data-center demand 61. Its market monitor estimated that data centers added $29.4 billion in capacity costs across the previous four auctions 61. Another account attributes $29.4 billion of additional electricity costs to AI data-center demand across PJM states 33. Data-center-driven transmission projects added more than $4.3 billion to the rate base paid by existing customers in 2024, according to the Union of Concerned Scientists 33.
Georgia Power’s Ashley Park–Wansley transmission project is expected to serve data centers with 70–80% of its projected electricity 25. This creates a tension between claims of broad public necessity and the project’s concentrated end use 25. The same distinction matters for Alphabet: a project may be legally classified as shared infrastructure while its economic rationale remains heavily dependent on a small number of hyperscale customers.
Opportunity and reliability risk
The buildout should benefit utilities, transmission developers, electrical-equipment suppliers, storage providers, and grid-software companies 25,46,58,59. Eaton identified data centers as a strong demand area in the second quarter of 2026 48. Civilian nuclear generation and small modular reactors are also being positioned as potential supply sources for hyperscale AI campuses 28,63, while large technology companies are acquiring or reserving reliable nuclear power for privately operated facilities 41. Tesla is preparing for energy-storage demand from data centers and broader electrification 64.
The opportunity does not remove the engineering problem. Alphabet’s growth could be constrained if power availability, energy-supply costs, or grid reliability limit the pace, location, and economics of new facilities 16. Large facilities connecting to or disconnecting from the grid create additional operational and reliability challenges 18, and concentrated dependence on a small number of grids is itself a risk 59,61. Is this truly negligible, or have we missed a coupling between facility expansion and system stability? The prudent answer is to assume that the coupling exists until field data proves otherwise.
Cost Allocation and the Social License to Operate
Developers are increasingly expected to pay
The political question is simple, though the accounting is not: who pays for the infrastructure required by hyperscale loads? Texas proposals would require data centers drawing at least 100 MW to cover the transmission and generation upgrades they demand, rather than spreading those costs across ratepayers 8. Related policy would require data centers to pay their share of grid infrastructure costs 24, reflecting concerns that rapid development could increase household bills, challenge reliability, and transfer costs to ordinary customers 24. The financial burden of electricity supply and supporting grid assets may increasingly fall directly on developers 33.
This arrangement may improve the economics of utilities and equipment vendors, but it raises the all-in cost of Alphabet’s capacity expansion. It may also favor sites with existing infrastructure, available grid stiffness, or long-term power contracts. In a power system, the cost of a new load is not confined to the meter. It includes the bridge, the approach roads, and sometimes the reinforcement of the bridge behind it.
The evidence is not uniform across locations. Counties with their own electrical grids experienced an average electricity-price increase of approximately 5% after data-center entry 47. The increase was statistically significant but substantially below claims that AI data centers could double or triple prices 47, although larger local effects remain possible 47. If grid capacity cannot expand with demand, other electricity users may bear higher prices 24,47, and data centers can contribute to local power-price increases 60. The appropriate conclusion is therefore a localized, material, and highly variable impact—not a uniform national price shock.
Alphabet’s social license will depend heavily on whether it funds incremental infrastructure and demonstrates that local benefits exceed local costs. Data centers can generate jobs and tax revenue 15, but their economic benefits are generally positive yet modest 47, more pronounced in dense areas with existing infrastructure 47, and dependent on local density, grid capacity, construction capability, and complementary businesses 47. Benefits vary substantially by location 47 and may amplify over time 47, while electricity, water, pollution, and infrastructure costs may fall partly on local communities 47.
Environmental, Permitting, and Community Constraints
Water, energy, and emissions
Environmental considerations are no longer peripheral to data-center strategy. Operators face growing pressure around decarbonization and sustainability compliance 70, including broader metrics such as energy reuse and Scope 3 emissions accounting 70. Energy efficiency has become a business-critical priority 62. AI data centers consume substantial electricity and water, can raise local utility costs, provoke opposition, and create ESG or regulatory liabilities 12. Across the evidence base, water, electricity, land use, and grid crowding appear as principal ESG issues 60, with related regulatory, community-relations, and operating-cost consequences 60.
Water intensity varies with cooling design, local climate, and technology choices 66. The burden is therefore not uniform, but neither is it trivial. Current data-center water consumption is described as approximately 6.6 million liters per day 27, with demand forecast to triple by 2030 27. Water demand may compete with public drinking-water supplies 26,27, imposing operational and regulatory constraints on expansion 27. Resource use can undermine public acceptance or raise operating costs 53, and facilities may worsen local resource constraints 53. The unresolved policy problem is how to measure and mitigate water, energy, pollution, and grid costs while preserving economic spillovers 47.
Permitting and political resistance
Data-center development is becoming a U.S. political and regulatory issue 22. Proposed state and local bans 22, as well as opposition to facilities already under consideration 10, show that permitting is becoming a gating factor rather than a clerical step. Local governments are reconsidering zoning, land-use frameworks, electricity demand, and long-term community planning 72, while some jurisdictions have revised zoning or imposed additional planning requirements 72. Large developments require major land-use decisions and can shape long-term community development 72.
Public protests and permitting resistance are recurring risks 53. A proposed 10-GW project has raised questions about power consumption, carbon intensity, grid capacity, water, permitting, and community impact 23. U.S. government control over power access is also a factor in at least one proposed project 23. Proposed bans therefore represent more than headline risk: they introduce political, execution, infrastructure-capacity, and regulatory uncertainty 22.
Choosing the power architecture
The strategic response is a choice among grid-supplied clean energy, firm nuclear or fossil generation, self-supply, and flexible load management. One policy debate contrasts a fossil-fuel-heavy approach with a diversified system combining renewables, storage, efficiency, demand response, advanced transmission, virtual power plants, and existing thermal and nuclear assets 59. Clean-energy deployment is characterized as the least-cost, lowest-risk response to uncertain demand 59, and clean energy could meet demand quickly enough 59. At only 33% realization of projected data-center demand, one analysis estimates that clean energy would save $2.6 billion relative to a fossil-oriented system 59.
The converse risk is equally important. Overbuilding fossil generation could leave consumers paying for unnecessary infrastructure if demand fails to materialize 59, creating stranded-asset risk for utilities and investors 59. Islanded data-center power plants could provide a self-supply pathway, but may externalize environmental costs 61. Alphabet’s durable advantage may therefore depend on combining procurement scale with low-carbon power, water-efficient cooling, demand flexibility, and credible community engagement.
Construction, Labor, and Delivery Capacity
Data-center construction is constrained primarily by power and cooling rather than physical floor space 55. Facilities are located according to infrastructure, skilled labor, and practical considerations—not latency alone 55. Construction demand is rising alongside labor costs 19, particularly for electricians, carpenters, and other skilled trades 19,20. Labor shortages and rapidly rising wages are already reported 19,20, while construction activity is concentrated in particular regional labor markets 19, creating upward pressure on wages and costs 19.
Data-center development can divert workers from housing and manufacturing 19, and staffing agencies and temporary labor are important to delivery capacity 19. Labor scarcity, compensation inflation, and construction delays can increase capital expenditure and defer revenue recognition 14,19. Grid-expansion delays are an explicit infrastructure risk 21, while electricity availability, energy prices, and reliability may determine project location and economics 16.
The effect is double-edged. Construction-related suppliers and utilities benefit from the buildout 58, but labor demand may be temporary rather than structural 20. Workers could face unemployment if construction slows 19, and companies exposed to the capital-spending cycle may experience volatile earnings 19. Project delays or cancellations have already been attributed to supply-chain and grid-connection constraints 59. A well-designed project plan must therefore account for the transient response of the labor market, not merely its peak load.
Demand, Financing, and Obsolescence Risk
The danger of collective overbuilding
The bullish narrative contains a necessary warning: capacity may be overbuilt. The rush to develop facilities increases the risk of overpayment and excess supply 57, and simultaneous overbuilding could turn data centers into a high-capital, low-return commodity business 57. Other claims identify potential collective overbuilding of data-center and compute capacity 49, possible construction overbuilding 20, and underutilization risk 5.
Physical infrastructure may depreciate rapidly or become obsolete as facilities are continually refreshed 56. Spending may fail to generate returns sufficient to cover depreciation, financing, and operating costs 56. Project-financed facilities can leave users responsible for rent or asset-value losses if demand falls short 44, while credit-rating agencies have warned about capacity financed through off-balance-sheet structures 71. Liquidity, financing, and credit-rating stress are therefore meaningful sector risks 14,53.
The financial exposure matters directly to Alphabet as a potential anchor tenant, customer, or developer of AI capacity. Data-center construction depends on GPUs, TPUs, equipment financing, energy, facilities, and long-term customer contracts 44. Continued buildouts by Amazon and other infrastructure providers may reduce compute costs over the medium term 43, potentially improving Alphabet’s AI economics while intensifying price competition. Rapidly evolving hardware may require facilities designed for frequent technology changes 52, and rack-scale systems are becoming an important infrastructure trend 35. Excess or obsolete capacity could pressure returns on invested capital even if aggregate AI demand remains strong.
The isolated claim that data centers need $200 billion in annual revenue to break even 11 should not be treated as a consensus estimate because it has only one source. It nevertheless illustrates the unusually high fixed-cost burden of the model.
Global Expansion and Location Strategy
The infrastructure challenge is global. China’s data centers consumed 166 billion kWh in 2024 65 and generated 85.9 million metric tons of carbon emissions 65, although electricity-consumption growth was expected to slow to 10.7% in 2025 65. India’s data-center capacity is projected to rise from 2.2 GW to 12 GW by 2030 67, and Australia’s market is accelerating 7. Australian facilities face high water and electricity requirements 54, with expansion constrained by energy availability, water, grid infrastructure, approvals, and community acceptance 54. Thailand faces rising electricity demand and shortages of skilled labor 46.
Southeast Asian location decisions are expected to reflect power costs, geopolitical and regulatory conditions, political stability, connectivity, labor availability, and sustainable operations 46. Israel has received approximately 27 GW of connection requests—roughly three times average national consumption and above its historical peak 29. China is emphasizing renewable-energy and surplus-clean-energy regions in data-center placement 41. Abundant power is an attraction for Texas 30, yet grid and permitting constraints can offset that advantage.
For Alphabet, geographic diversification broadens the opportunity but complicates location strategy. The global market brings exposure to energy-price volatility, geopolitical risk, regulatory divergence, and regional labor conditions. Data centers are becoming centers of innovation, economic power, and national influence because computational capacity is geographically concentrated 69. That concentration may increase the strategic value of Alphabet’s infrastructure while inviting greater government scrutiny.
Implications for Alphabet
Alphabet’s AI and cloud ambitions depend on physical infrastructure that cannot be scaled through software investment alone. The company’s competitive position benefits from access to compute, power, cooling, networking, and high-quality facilities, but those requirements raise capital intensity and execution risk. Data centers require extraordinary quantities of electricity, land, water, chips, cooling systems, transmission infrastructure, financing, and time 42. Operators also depend on accurate software visibility across mission-critical assets, power, cooling, and environmental systems 45. Unused or obsolete servers can continue consuming electricity and occupying rack space 45.
These conditions create openings for energy-efficiency, infrastructure-monitoring, cooling, storage, and flexible-load technologies 36,61. Alphabet’s likely priorities are to secure firm and diversified power, improve energy and water efficiency, align capacity additions with contracted demand, and reduce exposure to local political opposition. Data centers may need to generate their own power, shed load, shift timing, or participate as flexible grid resources 61. A portfolio combining renewable procurement, storage, demand response, nuclear or other firm low-carbon generation, and upgraded transmission is more robust than dependence on a single technology or grid. It also addresses the risk that traditional facilities depend on fossil-fuel-heavy grids 31.
The opportunity extends beyond Alphabet’s own facilities. Mining and metals demand is supported by AI data centers, grids, renewables, electrification, and resilient supply chains 13, and data-center investment is a potential catalyst for mining and metals 38. Companies supplying power, materials, and construction infrastructure may capture value even when individual data-center projects are delayed. Investors should, however, distinguish durable infrastructure demand from speculative capacity buildout.
Several lower-confidence claims should be discounted. The assertion that data centers are currently largely related to military usage 9 is weakly corroborated and inconsistent with the broader evidence describing hyperscale commercial, cloud, industrial, and civilian demand. Military compute investment is nevertheless a growing niche: the U.S. defense establishment is pursuing large-scale compute capacity 17, while military data-center contractors face permitting, cost, execution, social-license, regulatory, and demand risks 40, alongside environmental and pollution concerns 40. Similarly, claims that data centers are driving up prices of phones, laptops, and game consoles 39 and may consume roughly 70% of memory output 34 are isolated and should not be treated as established sector-wide conclusions.
Practical Conclusions
The evidence supports a constructive but selective view. Data-center demand is a credible secular driver for electricity generation, grid investment, storage, cooling, construction, and materials, but its magnitude and timing remain uncertain. The principal risks to Alphabet are not limited to insufficient AI demand. They include power-access delays, higher infrastructure costs, local opposition, regulatory burdens, labor scarcity, energy-price volatility 51, cyberattacks against concentrated infrastructure 69, and a mismatch between capacity investment and realized customer demand.
Proposed bans and planning restrictions 22 demonstrate that social license is becoming a gating factor. The possibility that cloud giants are having difficulty building facilities creates a supply constraint 50. That constraint could support pricing and preserve the value of Alphabet’s existing infrastructure, but it could also slow model deployment and raise costs.
The most defensible operating Ansatz is disciplined flexibility: secure firm and diversified power; prioritize energy- and water-efficient designs; fund incremental grid requirements; maintain geographic diversity; use storage and demand response where technically sound; and match capacity additions to contracted or otherwise well-supported demand. Forecast dispersion warrants particular caution. Estimates range from 78 GW to 194 GW of U.S. data-center power demand by 2035 1,37. Alphabet should therefore favor flexible, contracted, energy-efficient capacity over speculative buildout. In electrical engineering, as in finance, the most expensive failure is often not insufficient ambition but an elegant design built for a load that never arrives.