DGCP™ Analyst Article
When AI Capacity Became Dependent on Contracted Power Access
Date: 2026-09-09 (Asia/Bangkok)
Category: Analyst Article
Framework: DGCP™ — Data Governance & Continuous Proof
Mode: Observation • Structural Analysis • Evidence Context • No Prediction • No Advice
Location: Earth System
Observation
AI capacity can become dependent on the ability to secure power access, not simply on the existence of electricity somewhere in the system.
Recent agreements between technology companies, utilities, and electricity producers connect data-centre growth with power arrangements extending years or decades into the future. Some support continued output from operating plants. Others support restarts or generation projects that are not yet delivering electricity.
The agreements show that power access has entered the planning horizon of compute expansion. They do not show that every contracted megawatt has already been delivered or that every associated data centre is energized and operating.
Core Question
What changes when AI capacity depends not only on the existence of electricity or grid infrastructure, but on securing power access at the scale, duration, location, and delivery conditions required for compute expansion?
The question moves beyond regional generation totals. It asks what instrument was signed, what power is covered, where that power enters the system, when delivery is scheduled to begin, and which physical or regulatory steps remain incomplete.
Evidence Context
The International Energy Agency estimated that data centres consumed around 415 terawatt-hours of electricity globally in 2024. Its Base Case projected around 945 terawatt-hours in 2030 and identified AI as the most important driver of the increase. Total data-centre demand still includes AI, cloud, enterprise, storage, networking, and supporting infrastructure, so the total is not an AI-only measure. [1]
The IEA also estimated that grid constraints could delay around 20% of global data-centre capacity planned for construction by 2030. That estimate came from a location-specific analysis of congestion, grid policies, connection timelines, and upcoming data-centre projects. It was not an observed count of completed facilities already waiting for power. [2]
The same IEA work noted that building new transmission lines can take four to eight years in advanced economies, while critical grid-component lead times have lengthened materially. [1]
This article therefore examines contracted access using company, utility, regulatory, and project records. It separates operating generation from future generation, signed agreements from delivered electricity, and data-centre support from confirmed AI-specific compute output.
Evidence was reviewed through 2026-09-09. Future dates remain scheduled, expected, or planned unless a source documents operation.
Contracted Power Access
The Instrument Matters
A power purchase agreement, utility supply arrangement, interconnection agreement, capacity reservation, and infrastructure-development agreement do not create the same rights or obligations.
A corporate arrangement can support generation on a regional grid without creating a dedicated physical line from one plant to one data centre.
- Power generation ≠ site-level power access.
- Grid connection ≠ contracted electricity.
- Contracted electricity ≠ delivered electricity.
- Power agreement ≠ completed infrastructure.
- Delivered power ≠ operational or fully utilized compute.
Contract duration can support long-range planning, but it does not guarantee uninterrupted physical delivery. Grid operation, outages, transmission conditions, and contract terms remain relevant.
Contracted Volume Must Retain Its Status
A megawatt figure may describe existing output, future generation, a development ceiling, an incremental uprate, or a contracted quantity. Those states cannot be combined into one operating-capacity number.
The same discipline applies to compute. Announced data-centre load, initial energized load, and full buildout load are different states. A power agreement can support a future expansion plan without proving that the corresponding processors, buildings, networks, or cooling systems are complete.
Physical Delivery and Grid Reinforcement
Contracted power becomes operationally relevant only through the infrastructure and approvals that make delivery possible. Depending on the project, these may include generation construction, plant restoration, regulatory authorization, interconnection work, transmission or distribution upgrades, substations, transformers, commissioning, and energization.
The IEA's 2025 transmission-grid work reported procurement times of two to three years for cables and up to four years for large power transformers. Average lead times for those components had almost doubled since 2021. [3]
These findings do not establish that every contracted project requires new transmission or faces an equipment delay. They show why a signed electricity agreement cannot be treated as completed delivery without project-specific implementation evidence.
Selected Evidence Cases
Google, Kairos Power, and TVA: Scheduled First Delivery
Google and Kairos Power announced a master development agreement in October 2024 for a series of advanced nuclear projects totaling up to 500 MW by 2035. Kairos Power stated that it would develop, construct, and operate the projects, with the first deployment targeted for 2030. [4]
In August 2025, Google, Kairos Power, and the Tennessee Valley Authority identified Hermes 2 as the first deployment under that wider arrangement. A binding power purchase agreement between Kairos Power and TVA covers up to 50 MW from Hermes 2, with electricity intended for the TVA grid that serves Google data centres in Tennessee and Alabama. Google stated that the arrangement would help meet its data-centre electricity demand starting in 2030. [5]
The case directly links digital expansion with a defined power arrangement. It does not establish delivered power today. Hermes 2 remains a future electricity source, and the wider 500 MW figure remains a multi-project development target through 2035 rather than current operating generation or energized compute capacity.
Microsoft and Constellation: Contracted Restart With Remaining Delivery Conditions
Constellation and Microsoft signed a 20-year power purchase agreement in September 2024 covering the output of the proposed restart of Three Mile Island Unit 1, renamed the Crane Clean Energy Center. The renewed plant was described as approximately 835 MW, with Microsoft purchasing energy to help match the electricity used by its data centres in the PJM region with carbon-free energy. [6]
By August 2026, Constellation reported that FERC had approved a waiver allowing the transfer of existing Capacity Interconnection Rights to Crane and that the NRC had approved a fuel-license amendment. Constellation described both as major restart milestones and continued to target restart in 2027. [7]
The latest public implementation evidence therefore shows material regulatory progress, but not completed operating generation. Constellation's June 2026 filing also stated that the restart and delivery under the PPA remained subject to regulatory approvals and that PJM studies had identified transmission upgrades relevant to full deliverability. [8]
The signed contract is therefore not equivalent to operating power. It creates a defined commercial pathway while restart work, grid deliverability, remaining approvals, and commissioning continue to determine when electricity can flow under the arrangement.
Meta and Clinton: Contracting Around Existing Generation
Meta and Constellation announced a 20-year agreement beginning in June 2027 for the output of the operating Clinton Clean Energy Center. Meta stated that the arrangement secures 1,121 MW of existing emissions-free nuclear generation and supports an additional 30 MW through plant uprates. [9]
Constellation reported that the 30 MW uprates were expected to be fully complete in 2029. [10]
This case provides important counter-evidence. Contracted power access does not always require a new generating plant or a full new delivery system. A long-term agreement can support continued access to electricity from an operating asset. The incremental 30 MW nevertheless remains future capacity until the uprate is completed.
Counter-Evidence
The evidence does not support a universal claim that AI capacity requires long-term contracted power.
- Existing energized facilities can procure electricity through established utility arrangements and wholesale or retail markets without a new project-specific long-term generation contract.
- The Clinton case concerns an operating plant, showing that contracting can support continuity rather than create all underlying capacity from the beginning. [9] [10]
- On-site generation, storage, and demand flexibility can alter how a data centre obtains or manages power, although they do not automatically eliminate grid dependence. [2]
- The IEA identifies location in areas with stronger power and grid availability as another way to reduce connection-delay risk. [2]
- A contract can concern electricity delivered through a regional grid rather than a dedicated physical supply path to one facility.
Contracted access is therefore one observed mechanism within a wider power system. Its importance must be assessed at the level of the actual project, instrument, location, and delivery status.
What the Evidence Does Not Establish
The reviewed evidence does not establish that all AI capacity depends on long-term power contracts.
It does not prove that a larger contracted volume produces more operational compute or that companies with more agreements will achieve better commercial outcomes.
It does not convert scheduled generation into delivered electricity, or delivered electricity into installed processors, operational compute, or utilized AI services.
It does not show that every grid upgrade, transformer, substation, or interconnection requirement associated with these projects has been completed.
Corporate statements describe intended relationships among electricity procurement, data centres, and AI development. They do not establish the precise electricity consumption of individual AI workloads unless that measurement is separately reported.
No strategic advantage, market leadership, investment conclusion, or future competitive outcome is inferred from the agreements.
Closing Observation
The relevant power question for compute expansion is no longer only whether electricity exists within a region.
It is whether the required power has been contractually arranged, physically connected, and made deliverable at the relevant site, scale, and time.
AI capacity becomes physically meaningful only when the required power can be delivered to the facility at the scale and timing needed for operation.
A contract can create a pathway to that condition. It cannot substitute for the generation, grid, equipment, approvals, and commissioning that turn an agreement into operating power.
Evidence Discipline
This article separates data-centre electricity demand from AI-only demand, contract execution from electricity delivery, existing generation from future capacity, plant uprates from completed output, and regional-grid supply from dedicated site-level power.
Scheduled dates remain schedules. Development ceilings remain potential capacity. Corporate descriptions of intended use remain attributed company statements rather than independent measurements of operating AI compute.
For the Crane case, later 2026 implementation evidence supersedes earlier planning language where the two differ. Regulatory approvals already received are not described as pending, while remaining restart and deliverability conditions are preserved.
No claim predicts future project completion, AI dominance, electricity prices, or investment performance. The article provides no advice and makes no assessment reserved for SINTELRA strategic analysis.
Sources
[1] International Energy Agency. Energy and AI. 2025.
https://www.iea.org/reports/energy-and-ai
[2] International Energy Agency. AI and Energy Security. 2025.
https://www.iea.org/reports/energy-and-ai/ai-and-energy-security
[3] International Energy Agency. Building the Future Transmission Grid. 2025-02-25.
https://www.iea.org/reports/building-the-future-transmission-grid
[4] Kairos Power. Google and Kairos Power Partner to Deploy 500 MW of Clean Electricity Generation. 2024-10-14.
https://www.kairospower.com/updates/google-and-kairos-power-partner-to-deploy-500-mw-of-clean-electricity-generation
[5] Google. Our First Advanced Nuclear Reactor Project with Kairos Power and Tennessee Valley Authority. 2025-08-18.
https://blog.google/company-news/outreach-and-initiatives/sustainability/google-first-advanced-nuclear-reactor-project-with-kairos-power-and-tennessee-valley-authority/
[6] Constellation Energy. Constellation to Launch Crane Clean Energy Center, Restoring Jobs and Carbon-Free Power to the Grid. 2024-09-20.
https://www.constellationenergy.com/news/2024/Constellation-to-Launch-Crane-Clean-Energy-Center-Restoring-Jobs-and-Carbon-Free-Power-to-The-Grid.html
[7] Constellation Energy Corporation. Constellation Reports Second Quarter 2026 Results. 2026-08-06.
https://investors.constellationenergy.com/news-releases/news-release-details/constellation-reports-second-quarter-2026-results
[8] Constellation Energy Corporation. Form 10-Q for the Quarter Ended June 30, 2026. 2026.
https://www.sec.gov/Archives/edgar/data/1168165/000186827526000104/ceg-20260630.htm
[9] Meta. Meta and Constellation Partner on Clean Energy Project. 2025-06-03.
https://about.fb.com/news/2025/06/meta-constellation-partner-clean-energy-project/
[10] Constellation Energy Corporation. Form 10-Q for the Quarter Ended September 30, 2025. 2025.
https://www.sec.gov/Archives/edgar/data/1868275/000186827525000092/ceg-20250930.htm
Framework Notice
This public article follows the DGCP™ framework for observation-led structural analysis. It presents verified public evidence, explicit distinctions, counter-evidence, and known limitations.
It does not disclose internal DGCP™ scoring, thresholds, source weighting, source-ranking logic, comparison matrices, validation rules, decision rules, analytical sequences, workflow, proprietary methodology, governance architecture, or reconstructable strategic leverage models.
Observation Only • Structural Analysis • No Prediction • No Advice
Author
P’Toh
System Architect — DGCP™
License
DGCP | MMFARM-POL-2025
This work is licensed for public reading, citation, and reference with attribution to the author and framework.
Commercial reuse, modification, dataset extraction, model training, republication as another work, or removal of attribution requires prior written permission.