The data-center boom is driving a historic power-planning issue for electric utilities. The Electric Power Research Institute, Inc. (EPRI)¹ projects that data centers could account for 9% to 17% of U.S. electricity generation by 2030, and that growth is forcing utilities, regulators, and customers to decide how new power supply and grid upgrades should be paid for.

The Trump Administration has recently brought that cost question to the forefront. Through the Ratepayer Protection Pledge, it has said that major technology companies and hyperscale data-center developers should build, bring, or buy the power they need and pay the full cost of the related energy and grid infrastructure.

For utility stakeholders,² the practical issue is straightforward: data-center growth can support economic development, but it can also create rate pressure if new costs are spread across customers who do not directly benefit from the new load.

This white paper explains why new firm power will be expensive and why clear cost allocation will be central to utility planning.

The Outlook: New Megawatts Required to Serve Data Center Load

The U.S. data-center buildout is an industrial-scale electricity planning problem, not merely a real-estate or developing technology story. The scope of projected demand growth is unprecedented and presents fundamental questions about resource adequacy, cost allocation, and infrastructure investment that demand rigorous analysis by utility executives, their counsel, and regulators.

In June 2026, the Lawrence Berkeley National Laboratory (LBNL) updated its projections for U.S. data-center electricity consumption, concluding that these facilities could require between 521 and 843 terawatt-hours (TWh) annually by 2030, with a reference-case estimate of 649 TWh.³ To contextualize this figure: the reference case alone would represent approximately 11.8% of total U.S. electricity generation. Separately, the Electric Power Research Institute (EPRI) published a 2026 scenario range of approximately 380 to 790 TWh by 2030.⁴

Figure 1: 2030 Data-Center Electricity Forecasts — Implied Average Demand

LBNL further estimates that data centers consumed approximately 4.4% of U.S. electricity in 2023 (roughly 176 TWh), and that this share could rise to between 6.7% and 12% by 2028, corresponding to approximately 325 to 580 TWh.⁵ The U.S. Energy Information Administration’s January 2026 forecast identified the strongest four-year U.S. power-demand growth projection since 2000, driven largely by large computing facilities including data centers.⁶

Figure 2: Data Centers’ Share of U.S. Electricty Over Time (%)

In terms of peak-load capacity, EPRI projects aggregate data-center peak demand rising from approximately 21-22 gigawatts (GW) in 2024 to between 45 and 94 GW by 2030. PJM Interconnection’s 2025 long-term forecast projects 32 GW of peak-load growth between 2024 and 2030, with approximately 30 GW attributable to data centers.⁷ These projections imply significant upward pricing pressure and resource-adequacy concerns within the nation’s largest regional transmission organization.

Figure 3: EPRI Data Center Peak Load Growth (GW)

The International Energy Agency (IEA) has observed that data centers can be constructed in two to three years, while the power plants, transmission lines, substations, and interconnection upgrades required to serve them take considerably longer.⁸ This temporal mismatch is driving the trend toward “bring your own power” arrangements, co-located generation, and dedicated power purchase agreements (PPAs). The key challenge confronting developers, utilities, and regulators alike is securing firm, deliverable, hourly power at specific grid locations.

Why Data Centers Are Uniquely Difficult Electric Loads to Serve

A large artificial-intelligence data center does not behave like an office building, a hospital, or even a traditional manufacturing facility. It operates more like a continuously running industrial plant, demanding enormous quantities of firm, high-quality power twenty-four hours a day, every day of the year, with virtually zero tolerance for interruption.

Reliability and Redundancy Requirements

The Uptime Institute’s widely referenced tier classification framework illustrates the severity of these requirements:

• Tier I facilities require uninterruptible power supplies (UPS), dedicated cooling, and engine generators
• Tier III facilities permit concurrent maintenance without shutdown
• Tier IV adds full fault tolerance and continuous cooling redundancy⁹

For hyperscale AI workloads, operators typically target Tier III or Tier IV equivalence, requiring multiple independent utility feeds, on-site generation backup, and N+1 or 2N redundancy in all critical power and cooling systems.

Power Distribution Within the Facility

Power consumed within a data center is divided primarily between computing equipment and the systems that support it. Servers, storage arrays, and networking equipment account for approximately half or more of total facility electricity use, while cooling systems consume approximately 38-40% of total data-center electricity.¹⁰ The remaining balance supports power distribution, lighting, and ancillary systems.

Figure 4: Where Data Power Goes

The AI Rack Density Challenge

The transition from traditional enterprise computing to AI-driven workloads has driven the escalation in power density at the rack level. Legacy enterprise data centers typically operated at 5 to 10 kilowatts (kW) per rack, while current AI training racks demand 40 to 100+ kW per rack. New purpose-built AI facilities with liquid cooling may require 50 to 120 kW per rack and higher.¹¹

Figure 5: AI Rack Density Escalation (kW/rack)

This 10- to 24-fold increase in per-rack power density compounds every downstream infrastructure challenge:

• Electrical distribution
• Cooling capacity
• Water consumption
• Grid interconnection must all scale proportionally

ASHRAE has published updated guidance for thermal management of these high-density environments, emphasizing the transition from traditional air cooling to direct liquid cooling (DLC) as rack densities exceed 40-50 kW.¹²

Water and Siting Constraints

Water consumption represents another significant and often underappreciated siting constraint. A 1,000 MW U.S. data center may directly consume water at a rate comparable to approximately 26,000 average households. This places data centers in direct competition with agricultural, municipal, and environmental water uses, which is a reality that increasingly informs permitting decisions, community acceptance, and long-term operational risk.

Summary of Operating Requirements

In summary, proposed hyperscale data centers require:

• 24/7 hourly firm capacity with high reliability and redundancy
• Power quality and grid stability (as addressed by NERC’s large-load interconnection initiative)
• Continuous cooling capacity and water/heat rejection capability
• Fast deployment timelines to meet commercial commitments

Reliable Fuel Sources and the Requirement for Firm Power

Data centers require “firm power,” that is, generation resources that can deliver electricity reliably and continuously regardless of weather, time of day, or season. Intermittent renewable resources alone cannot satisfy data-center load profiles without substantial supplementation.

The Intermittency Problem

Without battery storage, wind and solar photovoltaic generation are inherently variable and non-dispatchable. Wind turbines generate electricity only when sufficient wind is available; solar panels produce power only during daylight hours and at reduced output under cloud cover. Neither resource can independently guarantee the continuous, hour-by-hour power delivery that a data center requires.

The capacity factor¹³ of a generation resource, defined as its actual output over a period divided by its maximum possible output, provides a useful metric for comparing the firmness of different fuel types. As Figure 6 illustrates, the disparity between dispatchable and intermittent resources is substantial.¹⁴

Figure 6: Capacity Factor by Generation Type

Nuclear power operates at approximately 93% capacity factor, effectively baseload, while natural gas combined-cycle plants achieve roughly 57%. By contrast, onshore wind averages approximately 35% and solar photovoltaic approximately 25%. Gas peaker units, designed for peak-demand periods, operate at only about 12% capacity factor but provide critical on-demand flexibility.

Reliability as a Non-Negotiable Requirement

The consequences of power interruption to hyperscale data-center operations are severe. Recent high-profile outages at facilities operated by Amazon Web Services and Microsoft have demonstrated that even brief service interruptions can cascade across global cloud infrastructure, affecting millions of enterprise and consumer users. For AI training workloads specifically, a power interruption can require restarting computational jobs that may have been running for days or weeks, representing significant economic loss in both computing resources and time-to-market.

The Current U.S. Generation Mix

To understand why new power for data centers will be expensive, one must first understand the existing electric generation fleet. The United States currently operates a diverse portfolio of generation assets, many of which are 20 to 50 years old and have been substantially or fully depreciated. These legacy assets benefit from lower sunk capital costs, existing interconnection rights, and transmission infrastructure already included in utility rate bases.¹⁵

This existing fleet delivers electricity at a lower average cost of service than any comparably reliable new-build generation could achieve today, which is a fundamental economic reality that underlies this paper’s central thesis. The implications of this embedded-cost advantage are explored in detail below.

New Energy Options Being Explored for Data Centers

The data-center industry and its utility partners are pursuing a broad portfolio of generation options to meet projected demand. Each carries distinct advantages, limitations, cost profiles, and regulatory considerations. This section surveys the principal alternatives currently under development or active consideration.

Grid Power Supplemented by New Renewable PPAs

Solar and wind resources can be constructed faster than nuclear plants or major transmission projects. The EIA expects solar generation to provide the largest single increase in U.S. generation capacity during 2026-2027, with approximately 70 GW of new solar capacity anticipated.¹⁶ Clearway Energy has announced approximately 1.2 GW of carbon-free PPAs with Google, illustrating the scale of corporate renewable procurement.¹⁷ However, intermittency requires supplementation with batteries, firm backup contracts, or transmission access to dispatchable resources.

Solar-Plus-Storage and Long-Duration Energy Storage

Hybrid projects pairing solar with battery storage are gaining traction. Enbridge is constructing a 365 MW solar facility paired with 200 MW/1,600 MWh of battery storage (the “Cowboy Project”) dedicated to serving Meta’s data-center load.¹⁸ For longer-duration needs, Meta and Noon Energy have announced plans for up to 1 GW/100 GWh of storage rated for 100+ hours of discharge duration, which is a technology that, if commercialized at scale, could meaningfully address the multi-day intermittency gap.

Co-Located Natural Gas Generation

Chevron and Microsoft have announced Project Kilby in West Texas: an approximately 2.67 GW co-located natural-gas generation complex with first power targeted for approximately 2028.¹⁹ This approach provides dedicated firm power directly adjacent to the data-center load, bypassing transmission and other grid constraints, but raising questions about emissions and long-term fuel-price exposure.

Behind-the-Meter Microgrids

Cummins is supplying natural-gas generating sets to Circe Energy for a West Texas AI and high-performance computing campus planned for deployment between 2026 and 2030.²⁰ Behind-the-meter configurations allow operators to avoid certain transmission and distribution charges and regulatory requirements, though they may face separate permitting and emissions constraints.

Fuel Cells

Bloom Energy has expanded its deployment at Equinix facilities beyond 100 MW of installed fuel-cell capacity. Oracle’s Project Jupiter at the BorderPlex facility envisions up to 2.45 GW of Bloom Energy fuel cells.²¹ Solid-oxide fuel cells offer high electrical efficiency, modularity, and relatively low emissions, but at capital costs that remain elevated relative to conventional thermal generation.

Existing Nuclear Restarts and Uprates

Constellation Energy and Microsoft have executed a 20-year PPA to restart Three Mile Island Unit 1 as the Crane Clean Energy Center, providing approximately 835 MW of carbon-free baseload generation.²² Nuclear restarts leverage existing site infrastructure, interconnection, and licensing frameworks, though they require significant capital investment for refurbishment and relicensing.

Nuclear Small Modular Reactors (SMRs)

Google and Kairos Power have entered a PPA with the Tennessee Valley Authority (TVA) for the Hermes 2 reactor, targeting up to 50 MW by approximately 2030 with plans for up to 500 MW by 2035.²³ SMRs promise factory fabrication, shorter construction timelines, and enhanced passive safety features, but no commercial SMR has yet achieved grid-connected operation in the United States.

Geothermal

XGS Energy and Meta have announced a 150 MW advanced geothermal project in New Mexico featuring zero operational water consumption.²⁴ Enhanced geothermal systems (EGS) offer firm, carbon-free generation with minimal land footprint, though commercial-scale deployment remains nascent and site-specific.

Hydrogen Fuel-Cell Backup and Emerging Technologies

Caterpillar, Microsoft, and Ballard Power Systems have demonstrated a 1.5 MW hydrogen fuel-cell backup system simulating a 48-hour grid outage event.²⁵ While promising as a zero-emission backup alternative to diesel generators, hydrogen’s current cost structure, storage logistics, and supply-chain limitations constrain near-term deployment at scale.

Helion Energy has executed a PPA with Microsoft targeting 50+ MW of fusion-generated electricity by 2028. This arrangement remains speculative and first-of-kind; fusion energy is not considered to be bankable under current commercial lending standards.

The Ingredient List for a New Power Facility

Constructing a new power generation facility to serve data-center load requires assembling multiple costly and constrained inputs. Using a natural-gas combined-cycle (CCGT) facility as a reference case, currently the most commercially proven option for firm dispatchable generation at scale, the essential components include:

Real Estate and Cooling Water Access

A suitable site must provide access to adequate cooling water (or dry-cooling alternatives at an efficiency penalty), sufficient acreage for the power block and associated infrastructure, and proximity to both natural-gas pipeline interconnection and high-voltage electrical transmission. Sites with existing cooling-lake access or river-water rights command premium valuations.

Power Generation Hardware

A modern CCGT facility in the 1,000-1,200 MW class typically employs twin combustion turbines paired with heat-recovery steam generators (waste-heat boilers) driving a steam turbine. Current lead times for large frame combustion turbines (GE 7HA, Siemens SGT6-9000HL class) extend to 36-48 months²⁶ amid strong global demand for gas-turbine capacity.

Workforce

A facility of this scale requires a skilled operating workforce, maintenance personnel, and management team. Labor-market constraints in skilled trades (instrumentation, controls, high-voltage electrical, rotating equipment) are a material consideration in project development timelines and operating costs.

Capital Investment

Current overnight capital costs for new CCGT generation range from approximately $1,200/kW under favorable conditions to $2,500/kW or higher where turbine shortages, supply-chain disruptions, or site-specific challenges apply. For a 1,200 MW facility, this implies a capital requirement of $1.4 billion to $3.0 billion before financing costs.²⁷

Revenue Requirements: How Regulated Utilities Recover Costs

Before analyzing the specific cost of new generation, it is essential to understand the regulatory framework through which electric utilities recover their costs and earn a return on invested capital. The Revenue Requirement formula is the foundation of utility ratemaking and determines the total revenue that a utility must collect from its customers.

Figure 7: Revenue Requirement Formula

Operating Expenses

Operating expenses encompass ongoing day-to-day costs necessary to provide service: labor, fuel or purchased power, maintenance, administrative overhead, and insurance. These costs are recovered dollar-for-dollar in rates and do not include capital investments, which are recovered through the rate-base mechanism.

Depreciation, Return Of Capital Invested

Depreciation represents the systematic recovery of capital investment in long-lived assets over their regulatory useful lives. It provides internal funds for reinvestment and, critically, reduces the rate base over time. A fully depreciated asset contributes zero to the rate base since it generates no return for shareholders, but it continues to provide service to customers at only its operating cost. This is the economic mechanism by which older generating assets deliver electricity at lower cost than new construction.

Taxes

Income taxes and other taxes (property, franchise) are included in the revenue requirement so that the utility earns its authorized return on an after-tax basis. Tax normalization and flow-through treatment affect the timing but not the ultimate magnitude of tax recovery.

Rate Base, Capital Investment Required

The rate base represents the value of property used and useful in providing service on which the utility is authorized to earn a return. It typically includes net plant in service (original cost less accumulated depreciation), working capital, and certain regulatory assets. It excludes customer contributions and, in most jurisdictions, construction work in progress (CWIP) until the asset enters commercial service.

Rate of Return, Return On Capital Invested

The allowed rate of return is the regulator-approved return on rate base, reflecting the utility’s weighted average cost of capital: the embedded cost of long-term debt plus the allowed return on equity. This rate must be set high enough to attract capital investment on reasonable terms while protecting customers from excessive charges. For investor-owned utilities, typical allowed returns on equity currently range from approximately 9.5% to 11%.

Cost of New Generation: CCGT Analysis and Benchmarks

This section presents a detailed analysis of the levelized busbar cost of electricity from a new natural-gas combined-cycle generating facility — currently the most commercially viable option for firm, dispatchable, utility-scale generation — and compares it against published industry benchmarks.

Levelized Cost Benchmarks (Lazard 2025)

Lazard’s 2025 Levelized Cost of Energy (LCOE) analysis provides widely referenced benchmarks across generation technologies. As shown in Figure 8, the range of costs varies dramatically by technology and reflects the fundamental trade-off between cost and firmness.²⁸

Figure 8: Selected New-Build LCOE Ranges, 2026

It is critical to note that the wind cost range of $37-$86/MWh reflects energy-only costs and excludes the additional expense of storage, firming contracts, or backup generation required to deliver 24/7 power to a data center. When firming costs are included, the effective all-in cost of renewable-plus-storage solutions rises substantially, often approaching or exceeding the cost of new thermal generation.

New CCGT Busbar Cost Analysis

A new non-CCS natural-gas combined-cycle facility carries a base-case busbar cost of approximately 6.1 cents per kilowatt-hour (~$61/MWh).

Base-Case Build-Up:

• Overnight capital cost: ~$1,400/kW
• Return on Equity: 10%
• Cost of Debt: 6%
• Life: 30 years
• Fixed O&M: $20/kW-year
• Variable O&M: $3.75/MWh
• Heat rate: 6.4 MMBtu/MWh
• Delivered gas price: $4.00/MMBtu
• Capacity factor: 80%
• Income taxes: 21% (federal only)

Fuel cost is the single largest component of CCGT generation cost. Each $1/MMBtu change in delivered natural-gas price moves the all-in generation cost by approximately 0.64 cents per kWh ($6.40/MWh). For a 1,200 MW facility operating at base-case assumptions, the annual revenue requirement is approximately $515 million.

Important Caveats

Busbar generation cost is not a retail electric rate. It excludes transmission, distribution, riders, taxes, and legacy or policy costs layered into delivered rates. For reference, 2025 average U.S. retail electricity prices were approximately 13.41 cents/kWh for commercial customers and 8.62 cents/kWh for industrial customers.²⁹

A significant regulatory uncertainty affects the long-term cost outlook for gas-fired generation. The EPA’s April 2024 Section 111 rule would require new gas turbines operating above 40% capacity factor to capture 90% of CO2 emissions via carbon capture and sequestration (CCS) by 2032. EIA models a 2031 non-CCS CCGT LCOE of $77.46/MWh under this baseline rule, versus $55.51/MWh if the rule is repealed.³⁰ EPA proposed repeal of this rule in June 2025; the regulatory outcome remains uncertain as of this writing.

Embedded vs. Incremental Cost: The Core Hypothesis

The long-run incremental cost of supplying a new, firm data-center load with new dedicated generation is likely to be materially higher than the embedded average cost of the existing generation fleet. This cost differential arises from basic economic principles and is not attributable to any single actor, political party, policy failure, or market distortion.

Why Embedded Costs Are Lower

The existing generation fleet benefits from several cost advantages that new construction cannot replicate:

• Sunk and substantially depreciated capital: Many existing plants have recovered most or all of their original construction costs through decades of depreciation charges. Their rate-base contribution, and therefore their capital-cost recovery through rates, is correspondingly low.
• Existing interconnection rights: Current generators hold firm interconnection agreements and transmission service rights that would be costly and time-consuming to replicate for new facilities.
• Transmission already in rate base: The transmission infrastructure serving existing generators has itself been depreciated over time, contributing relatively little to current rates.

The Magnitude of the Cost Differential

Figure 9: Embedded vs. Incremental Cost of Power ($/MWh)

The practical significance of this cost differential becomes apparent at data-center scale. A 100 MW data center operating at 95% load factor consumes approximately 832,200 MWh per year.³¹ At this consumption level, every $10/MWh premium in generation cost translates to approximately $8.3 million per year in additional electricity expense. Over a 20-year facility life, this single cost differential compounds to $166 million over time.

Moreover, the flat load profile of a data center means that its demand coincides with system peak, the most expensive hours on the grid. This requires accredited capacity resources, planning reserves, transmission capacity, and standby generation, all of which carry costs that do not arise for interruptible or peak-shaving loads.

When the Hypothesis May Not Hold

Intellectual honesty requires acknowledging scenarios where new dedicated generation could approach or match embedded fleet costs:

• Where the existing fleet is old, inefficient, or carries high fuel costs (e.g., aging coal plants with expensive delivered coal)
• Where the data-center load is genuinely flexible or contractually interruptible, reducing the firmness premium
• Where genuine surplus generation capacity and transmission headroom exist, allowing incremental load to be served at marginal rather than fully allocated cost
• Where the relevant comparison is to market peak-hour prices rather than embedded fleet average cost

These exceptions, while real, do not negate the general principle. In most regions and most scenarios, new firm generation for dedicated data-center load will cost more than the average embedded cost of the existing system.

Conclusion and Key Takeaways

New power for data centers is likely to be more expensive than power from the existing fleet. The reason is straightforward: new plants, grid upgrades, equipment, and financing must be paid for at today’s costs, while much of the existing fleet was built and recovered over decades.³²

Summary and Conclusion Exhibit: Revenue Requirement for New Firm Data-Center Power

The base case assumes a 1,200 MW CCGT facility operating at an 80% capacity factor, producing 8,409,600 MWh annually. The exhibit applies the paper’s assumptions to show the annual revenue requirement and the equivalent cost per MWh.

Figure 10: Year-1 Revenue Requirement — Cost-of-Service (Rate-Base) Method

Annual revenue the plant must earn in its first operating year to recover all costs and the allowed return.

Based on the information reviewed, new electricity will cost more than produced by the US installed base given:

• Capital costs reflect inflation from previous years, increases electric rates
• Operating costs also reflect cumulative inflation, increases electric rates
• Interest rates are still higher than Federal reserve targets, increases electric rates
• Equity costs are derived from interest rates, still high, increases electric rates

New power for data centers is likely to be more expensive than power from the existing fleet as much of the existing fleet was built and at lower costs, recovered over decades.³³ While electricity from a new facility providing firm power will cost over $60/MWh, as compared to regional averages ranging from $30/MWh to $55/MWh.³⁴

Key Takeaways for Utility Stakeholders

Plan for the full cost of new load.

Large data-center projects can require new generation, transmission, and reliability investments. Utility decisions should identify who pays for those costs before capacity is built.

Power availability means more than energy supply.

For data centers, power must be available at the right location and in sufficient quantity with enough reliability, cooling, and backup capacity to support continuous operations.

Cost allocation is the central policy issue.

If utilities build too much new capacity to service speculative load, existing customers may face stranded costs. If utilities build too little, developers may face delays and reliability risks and price increases.

Strong project planning reduces risks for the grid and the community.

The most attractive projects will bring committed capital, dedicated power supply, practical water and cooling plans, and clear benefits for the host utility system and community.

1. Electric Power Research Institute (EPRI).
2. Investors and ratepayers.
3. Lawrence Berkeley National Laboratory, “United States Data Center Energy Usage Report,” June 2026. Reference case projects 649 TWh, representing 11.8% of projected total U.S. electricity generation.
4. Electric Power Research Institute (EPRI), “Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption,” 2026 Update.
5. Lawrence Berkeley National Laboratory, June 2026 Update. Data centers consumed approximately 176 TWh (4.4% of U.S. electricity) in 2023.
6. U.S. Energy Information Administration, Annual Energy Outlook, January 2026. Identifies strongest four-year demand growth forecast since 2000.
7. PJM Interconnection, 2025 Long-Term Load Forecast. Projects 32 GW peak-load growth 2024-2030, approximately 30 GW attributable to data centers.
8. International Energy Agency (IEA), “Electricity 2024: Analysis and Forecast to 2026,” January 2024.
9. Uptime Institute, “Tier Standard: Topology,” incorporating Tier I through Tier IV classifications.
10. U.S. Department of Energy; Uptime Institute annual surveys. Cooling loads typically represent 38-40% of total data-center electricity consumption.
11. ASHRAE Technical Committee 9.9, “Thermal Guidelines for Data Processing Environments,” 5th Edition. See also Uptime Institute, “Global Data Center Survey,” 2025.
12. ASHRAE (American Society of Heating, Refrigerating and Air-Conditioning Engineers), “Liquid Cooling Guidelines for Datacom Equipment Centers,” 2024 Update.
13. Capacity factor = Annual Generation / (Capacity × 8760).
14. U.S. Energy Information Administration, Electric Power Monthly, Table 6.07.B, “Capacity Factors for Utility Scale Generators.”
15. EIA, “Electric Power Annual 2024,” Table 4.1, average age of operating generators by fuel type.
16. U.S. Energy Information Administration, Short-Term Energy Outlook, 2026. Projects approximately 70 GW of new solar capacity additions in 2026-2027.
17. Clearway Energy Group announcement, approximately 1.2 GW of carbon-free power purchase agreements with Google LLC.
18. Enbridge Inc., “Cowboy Solar + Storage Project” announcement: 365 MW solar + 200 MW/1,600 MWh battery storage, contracted to Meta Platforms.
19. Chevron Corporation and Microsoft Corporation, “Project Kilby” announcement, West Texas. Approximately 2.67 GW of natural-gas generation, first power targeted ~2028.
20. Cummins Inc. announcement: natural-gas generating sets supplied to Circe Energy for AI/HPC campus in West Texas, 2026-2030 deployment.
21. Bloom Energy Corporation filings and announcements. Equinix deployment exceeds 100 MW; Oracle Project Jupiter at BorderPlex envisions up to 2.45 GW.
22. Constellation Energy and Microsoft Corporation, 20-year PPA for restart of Three Mile Island Unit 1 (“Crane Clean Energy Center”), approximately 835 MW.
23. Google LLC, Kairos Power, and Tennessee Valley Authority PPA for Hermes 2 reactor, up to 50 MW by ~2030, with plans for 500 MW by 2035.
24. XGS Energy and Meta Platforms, 150 MW advanced geothermal project in New Mexico with zero operational water consumption.
25. Caterpillar Inc., Microsoft Corporation, and Ballard Power Systems, 1.5 MW hydrogen fuel-cell backup demonstration simulating 48-hour grid outage.
26. “Siemens Energy CEO: Gas-fired power plant construction will be ‘photo finish,’” Welt, November 14, 2025.
27. EIA Annual Energy Outlook 2025, Table 8.2, “Cost and Performance Characteristics of New Generating Technologies.” EPRI generation cost estimates, 2025.
28. Lazard, “Levelized Cost of Energy Analysis,” Version 16.0, 2025. Unsubsidized ranges for new-build generation in the United States.
29. EIA, Electric Power Monthly, Table 5.6.A, “Average Retail Price of Electricity,” 2025 data. Commercial: 13.41 cents/kWh; Industrial: 8.62 cents/kWh.
30. EPA, “New Source Performance Standards for Greenhouse Gas Emissions From New, Modified, and Reconstructed Fossil Fuel-Fired Electric Generating Units,” 89 Fed. Reg. 33240 (April 2024). Proposed repeal published June 2025.
31. Calculation: 100 MW x 8,760 hours/year x 0.95 capacity factor = 832,200 MWh/year.
32. See EIA historical data on power-plant construction costs. Nominal costs have increased substantially since the 1990s across all fuel types, reflecting materials inflation, labor costs, environmental compliance, and supply-chain factors.
33. See EIA historical data on power-plant construction costs. Nominal costs have increased substantially since the 1990s across all fuel types, reflecting materials inflation, labor costs, environmental compliance, and supply-chain factors.
34. EIA.