Should AI bring its own electricity supply?

Introduction

The race to build AI infrastructure is increasingly colliding with a less glamorous constraint: electricity connections. In many regions, the biggest delay facing new AI data centres is not financing, chips, land or even total power supply. It is the wait for grid access. Utilities may need years to expand substations, transmission lines and transformers before a new facility can draw the hundreds of megawatts it requires. As a result, some developers are exploring a different approach: bringing their own electricity supply.[IEA]iea.orgGlobal data centre capacity additions in the Base Case and…3 Apr 2025 — Global data centre capacity additions in the Base Case and…

This shift matters because the larger vision of AI-enabled abundance depends on enormous growth in computing capacity. If advanced AI is to accelerate science, medicine, engineering and other forms of human progress, the underlying physical infrastructure must expand as well. Dedicated on-site power may help bypass grid bottlenecks, but it also raises difficult questions about emissions, energy planning, public costs and who benefits from scarce electricity infrastructure. The debate is not simply about how to power data centres. It is about whether AI’s infrastructure buildout will strengthen public energy systems or increasingly operate alongside them.

Why developers consider dedicated gas or local generation

The attraction of on-site power is straightforward. A developer can often build generation equipment faster than waiting for major grid upgrades.

Large AI facilities increasingly require power on the scale of industrial regions or small cities. In many electricity markets, connection queues stretch for years. Grid operators must conduct studies, approve upgrades and secure equipment before new loads can connect. Transformer shortages and transmission constraints add further delays. The International Energy Agency has warned that a significant share of planned data-centre capacity is at risk from grid bottlenecks.[IEA]iea.orgElectricity 2026 – AnalysisA lack of grid capacity is emerging as a critical bottleneck in many regions, driving higher levels of congest…[IEA]iea.orgEnergy supply for AINatural gas and coal together are expected to meet over 40% of the additional electricity demand from data centres…

For companies spending billions on AI hardware, delays can be extraordinarily expensive. Advanced chips depreciate rapidly in strategic value. A facility that opens three years late may miss key technology cycles. This creates strong incentives to secure power independently rather than wait for conventional utility timelines.

The most common proposal is “behind-the-meter” generation. Instead of relying entirely on the public grid, a data-centre campus builds power generation directly on site or adjacent to it. That generation can include:

• Natural-gas turbines.
• Combined heat and power systems.
• Large-scale battery storage.[utilitydive.com]utilitydive.comUtility Dive Utilities are spending billions on the data center boomMar 6, 2026 — The AI data center frenzy is shifting utilities' focus to large-scale generation. But advocates say flexible, distributed e…
• Solar installations paired with storage.
• Microgrids that can operate independently or alongside the wider grid.
• Future proposals involving small modular nuclear reactors, although these remain largely prospective rather than widely deployed.[DataCenterKnowledge]datacenterknowledge.comData Center Knowledge Why Data Centers Are Turning to Behind-the-Meter PowerWhy Data Centers Are Turning to Behind-the-Meter PowerApril 15, 2026 — 15 Apr 2026 — Grid queues, community pressures…Published: April 15, 2026

Natural gas has emerged as the dominant near-term option because it can provide continuous power. Unlike wind and solar generation alone, gas turbines can run around the clock and support the high utilisation rates that AI training clusters often require. The IEA reports a growing pipeline of on-site gas projects linked to US data centres, reflecting the urgency of securing reliable power quickly.[IEA]iea.orgPipeline of onsite natural gas power projects for data…Apr 16, 2026 — Pipeline of onsite natural gas power projects for data centre…

This is partly a consequence of how valuable computing capacity has become. If advanced AI systems continue improving rapidly, firms may view years-long grid delays as a larger business risk than the costs of constructing dedicated power infrastructure.

The appeal: faster growth and more reliable computing

Supporters of dedicated generation argue that it offers benefits beyond simply avoiding queues.

First, it can accelerate infrastructure deployment. New data centres may be able to begin operating before transmission upgrades are completed. In a world where AI capability depends heavily on available compute, this can significantly affect development timelines.[DataCenterKnowledge]datacenterknowledge.comData Center Knowledge Why Data Centers Are Turning to Behind-the-Meter PowerWhy Data Centers Are Turning to Behind-the-Meter PowerApril 15, 2026 — 15 Apr 2026 — Grid queues, community pressures…Published: April 15, 2026

Second, on-site generation can improve resilience. AI facilities increasingly support cloud services, research systems and critical digital infrastructure. Local generation can reduce exposure to transmission failures, regional shortages or grid instability.

Third, some advocates argue that private generation reduces pressure on existing customers. If technology firms finance their own generation assets rather than drawing entirely from public infrastructure, they may avoid competing directly for limited grid capacity. This logic appears in proposals that large AI projects should “pay their own way” through dedicated generation and grid investments.[Tom's Hardware]tomshardware.comThis move aligns with U.S. policy pressures, especially from former President Trump, requiring tech firms to self-fund infrastructure imp…

There is also a longer-term argument linked to the broader AI bloom thesis. If advanced AI eventually contributes to major gains in scientific discovery, medical research, clean-energy innovation and economic productivity, then delays in compute infrastructure may carry large opportunity costs. From this perspective, faster access to power is not merely a commercial advantage. It could affect how quickly society gains access to potentially transformative capabilities.

The challenge is that the same infrastructure choices that accelerate AI deployment can create new environmental and governance problems.

What on-site power changes for emissions and climate goals

The strongest criticism of dedicated generation is that it can lock in additional fossil-fuel use.

Many technology companies have publicly committed to clean-energy goals. Yet grid constraints are pushing some operators towards gas generation because it is available now and can provide constant power. The IEA expects natural gas and coal to supply a substantial share of additional electricity demand from data centres through 2030.[IEA]iea.orgovercoming energy constraints is key to delivering on europe s data centre goalsOvercoming energy constraints is key to delivering on…16 Nov 2025 — Expensive delays in core markets are prompting data centre inve…

Critics argue that this creates a mismatch between AI’s long-term promise and its near-term energy trajectory. If AI is expected to help accelerate climate science, energy innovation and environmental management, but its infrastructure expansion depends heavily on new fossil-fuel assets, the transition becomes more complicated.

Several concerns appear repeatedly:

• New gas plants can operate for decades, extending fossil-fuel dependence.
• Methane leakage throughout gas supply chains can significantly increase climate impacts.
• Behind-the-meter facilities may be less visible within traditional utility planning frameworks.
• Pressure for rapid deployment may encourage exemptions or weakened environmental review processes.[Grist]grist.orgdata centers natural gas methane behind the meterEarly boosters of the data center boom suggested…Read more… EarthRights Some reporting has highlighted proposals where emissions associated with planned gas-powered data-centre campuses could rival those of entire[earthrights.org]earthrights.orgai data center demand is fueling the pipeline industryEarthRights InternationalAI Data Center Demand is Fueling the Pipeline Industry28 Apr 2026 — 40% of data centers in the US are powered by… countries if replicated at scale. While estimates vary substantially depending on utilisation rates and technology choices, the broader concern is clear: dedicated power can reduce one bottleneck while creating another.[WIRED]wired.comNew Gas-Powered Data Centers Could Emit More…22 Apr 2026 — New gas projects linked to just 11 data center campuses around the US…

At the same time, the picture is not entirely one-sided. A grid connection does not automatically mean low-carbon electricity. In some regions, data centres drawing from the public grid are also consuming power generated by fossil fuels. The relevant comparison is therefore not between perfect renewables and on-site gas, but between different pathways for meeting rapidly growing demand.

Why private generation can complicate grid planning

Electricity systems work best when generation and transmission are planned together. Large-scale movement towards private power can make that harder.

Utilities typically invest in infrastructure based on forecasts of future demand. When large AI projects suddenly announce gigawatt-scale loads, planners must decide whether to build new transmission lines, generation assets and substations. If those data centres later switch to self-generation, demand forecasts can change dramatically.

This creates a coordination problem.

Utilities may invest billions preparing for loads that never fully materialise. Alternatively, they may underinvest if they assume developers will remain self-sufficient, only to discover later that those facilities still require substantial grid support. Analysts increasingly warn about uncertainty surrounding data-centre demand projections and the risks of building infrastructure based on forecasts that may later shift.[Utility Dive]utilitydive.comUtility Dive Utilities are spending billions on the data center boomMar 6, 2026 — The AI data center frenzy is shifting utilities' focus to large-scale generation. But advocates say flexible, distributed e…

Most “off-grid” projects are not truly isolated from the wider system. They often retain grid connections for backup power, trading electricity or reliability support. In practice, many facilities occupy a middle ground between complete independence and full utility dependence.[DataCenterKnowledge]datacenterknowledge.comData Center Knowledge Why Data Centers Are Turning to Behind-the-Meter PowerWhy Data Centers Are Turning to Behind-the-Meter PowerApril 15, 2026 — 15 Apr 2026 — Grid queues, community pressures…Published: April 15, 2026

This can create difficult questions about cost allocation:

• Who pays for transmission upgrades?
• Who funds reliability reserves?
• How should backup access be priced?
• What happens if anticipated data-centre demand never appears?
• Should ordinary consumers bear part of the infrastructure cost?[Utility Dive]utilitydive.comUtility Dive Utilities are spending billions on the data center boomMar 6, 2026 — The AI data center frenzy is shifting utilities' focus to large-scale generation. But advocates say flexible, distributed e…

These issues are increasingly becoming regulatory rather than purely technical debates.

How private power affects public access and local communities

The social impact of on-site power often depends on who captures the benefits and who absorbs the costs.

Supporters argue that private generation reduces strain on public infrastructure and can bring investment, jobs and local tax revenue. In regions struggling to attract industry, large AI campuses may be welcomed as major economic projects.

Opponents worry about a different outcome: the emergence of what some observers describe as a parallel energy system built primarily for the largest technology companies.[The Washington Post]washingtonpost.comto power their artificial intelligence infrastructure independently from traditional electricity grids. These projects, like the GW Ranch…

Several concerns drive local resistance:

Air pollution and health effects. Gas generation creates emissions beyond carbon dioxide, including nitrogen oxides and other pollutants. Communities located near industrial facilities may face cumulative environmental burdens.[Reuters]reuters.comAI data centers are forcing dirty 'peaker' power plants back into serviceThese plants, like the eight-unit petroleum-fired Fisk plant in Chicago, are designed to operate only during high demand but are now bein…

Land and infrastructure use. Large campuses can reshape local development patterns, require new pipelines or transmission infrastructure and increase pressure on water resources.

Electricity pricing. Residents often worry that utilities will expand infrastructure to support AI growth while spreading costs across ordinary customers. Regulators increasingly face demands to ensure large technology firms bear the costs associated with their own expansion.[Tom's Hardware]tomshardware.comThis move aligns with U.S. policy pressures, especially from former President Trump, requiring tech firms to self-fund infrastructure imp…

Political influence. Wealthy firms may secure faster access to power, equipment and permitting than smaller businesses or public institutions. Critics argue that this risks concentrating both computing capacity and energy access in the hands of a relatively small number of actors.

These concerns connect to a larger question running through debates about AI abundance. Even if advanced AI dramatically increases productive capacity, who gains access to the resulting benefits? Infrastructure decisions made during the buildout phase can influence whether future gains are broadly shared or concentrated.

Could data centres become flexible energy partners instead?

The choice may not be limited to either traditional grid dependence or permanent self-generation.

An emerging idea is to make AI facilities more flexible consumers of electricity. Certain workloads can potentially be shifted in time, paused temporarily or moved between locations when power systems are under stress. If successful, this could reduce the need for expensive infrastructure upgrades while allowing more facilities to connect sooner.

Field demonstrations have shown that some AI workloads can reduce electricity use during peak periods without severely affecting performance. Researchers and grid operators increasingly describe data centres as potential “grid-interactive” assets rather than purely passive consumers.[arXiv]arxiv.orgTurning AI Data Centers into Grid-Interactive Assets: Results from a Field Demonstration in Phoenix, ArizonaJuly 1, 2025…Published: July 1, 2025

This approach does not eliminate the need for more generation and transmission. However, it suggests that part of the bottleneck may be managed through smarter coordination rather than simply building more gas turbines.

In the longer term, developers are also exploring combinations of:

The practical challenge is timing. Grid expansion can take a decade. AI investment cycles often operate on timescales of months or a few years. Dedicated generation is partly a response to that mismatch.

The deeper question behind on-site power

The debate over dedicated electricity supply is ultimately a debate about how civilisation should build the infrastructure behind advanced AI.

One path prioritises speed. If AI systems can accelerate scientific discovery, medical progress and economic growth, then reducing compute bottlenecks may produce large social benefits. From that perspective, on-site generation is a pragmatic response to slow-moving energy systems.

Another path emphasises integration. Rather than building a parallel energy architecture around AI, it seeks to expand public grids, improve transmission, deploy cleaner generation and ensure infrastructure serves broader social goals alongside AI development.

In practice, most countries will likely end up somewhere between these models. Some dedicated generation will probably be built because grid delays are real and expensive. Yet even the largest technology firms generally still prefer reliable grid connections over complete energy independence. The IEA notes that on-site gas generation is growing, but it does not remove the underlying need to expand and modernise electricity networks.[theenergy.co]theenergy.coai data centres risk becoming victims of their own success iea warnsAI data centres risk becoming victims of their own success…Apr 19, 2026 — “Though uncertainties are high,” the IEA concludes, around 1…

That tension makes on-site power more than a technical workaround. It is an early test of how societies handle a recurring question in the age of advanced AI: when a potentially transformative technology collides with public infrastructure limits, should private actors build around the bottleneck, or should the bottleneck itself be fixed?

Amazon book picks

Further Reading

Books and field guides related to Should AI bring its own electricity supply?. Use these as the next step if you want deeper reading beyond the article.

Book

Human Compatible

By Stuart Russell

Explores how powerful AI could advance humanity if aligned with human values.

See on Amazon

Book

Superforecasting

By Philip Tetlock, Dan Gardner

Directly addresses prediction, uncertainty and decision errors.

See on Amazon

Book

The Coming Wave

By Mustafa Suleyman

Directly addresses whether advanced AI can deliver broad civilisational benefits while remaining governable.

See on Amazon

Book

The Better Angels of Our Nature

By Steven Pinker

Provides broad context for understanding patterns of violence.

See on Amazon

Browse more on Amazon:Human CompatibleSuperforecastingThe Coming Wave

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

genecon nakamura hand operated generator model b10-2631

SearcheBay.co.uk: generator model

Browse similar oneBay.co.uk

Example eBay listing

1/32 (and 1/35) Terex Lighting Generator (ID 13.08)

SearcheBay.co.uk: generator model

Browse similar oneBay.co.uk

Example eBay listing

1/32 (and 1/35) Kubota D1105 Engine Generator (ID 13.13)

SearcheBay.co.uk: generator model

Browse similar oneBay.co.uk

Example eBay listing

Model Smoke Generator USB powered ultrasonic railway layout war diorama dolls

SearcheBay.co.uk: generator model

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: iea.org
Link:https://www.iea.org/data-and-statistics/charts/global-data-centre-capacity-additions-in-the-base-case-and-capacity-at-risk-of-connection-delay-due-to-grid-constraints

2. Source: iea.org
Link:https://www.iea.org/reports/electricity-2026/grids

3. Source: datacenterknowledge.com
Title: Data Center Knowledge Why Data Centers Are Turning to Behind-the-Meter Power
Link:https://www.datacenterknowledge.com/energy-power-supply/why-data-centers-produce-their-own-power

Published: April 15, 2026

4. Source: iea.org
Link:https://www.iea.org/reports/energy-and-ai/energy-supply-for-ai

5. Source: iea.org
Link:https://www.iea.org/data-and-statistics/charts/pipeline-of-onsite-natural-gas-power-projects-for-data-centres-in-the-united-states-2026

6. Source: theenergy.co
Title: ai data centres risk becoming victims of their own success iea warns
Link:https://theenergy.co/article/ai-data-centres-risk-becoming-victims-of-their-own-success-iea-warns

7. Source: grist.org
Title: data centers natural gas methane behind the meter
Link:https://grist.org/energy/data-centers-natural-gas-methane-behind-the-meter/

8. Source: earthrights.org
Title: ai data center demand is fueling the pipeline industry
Link:https://earthrights.org/blog/ai-data-center-demand-is-fueling-the-pipeline-industry/

9. Source: wired.com
Link:https://www.wired.com/story/new-gas-powered-data-centers-could-emit-more-greenhouse-gases-than-entire-nations/

10. Source: iea.org
Title: overcoming energy constraints is key to delivering on europe s data centre goals
Link:https://www.iea.org/commentaries/overcoming-energy-constraints-is-key-to-delivering-on-europe-s-data-centre-goals

11. Source: reuters.com
Title: AI data centers are forcing dirty ‘peaker’ power plants back into service
Link:https://www.reuters.com/business/energy/ai-data-centers-are-forcing-obsolete-peaker-power-plants-back-into-service-2025-12-23/

12. Source: arxiv.org
Link:https://arxiv.org/abs/2507.00909

Published: July 1, 2025

13. Source: arxiv.org
Link:https://arxiv.org/abs/2605.03751

Published: May 5, 2026

14. Source: iea.org
Link:https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions

15. Source: iea.org
Link:https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

16. Source: iea.org
Link:https://www.iea.org/

17. Source: iea.org
Link:https://www.iea.org/reports/electricity-2026/executive-summary

18. Source: iea.org
Link:https://www.iea.org/reports/energy-and-ai/executive-summary

19. Source: datacentremagazine.com
Title: The IEA
Link:https://datacentremagazine.com/news/iea-can-our-power-grids-cope-with-the-age-of-electricity

20. Source: tomshardware.com
Link:https://www.tomshardware.com/tech-industry/artificial-[intelligence

21. Source: utilitydive.com
Title: Utility Dive Utilities are spending billions on the data center boom
Link:https://www.utilitydive.com/news/utilities-data-center-risk-buildout-bubble-ai-infrastructure-grid/812781/

22. Source: washingtonpost.com
Link:https://www.washingtonpost.com/business/2026/02/19/data-centers-power-grid-ai/

23. Source: utilitydive.com
Title: gas power plant boom bust data center ai trump
Link:https://www.utilitydive.com/news/gas-power-plant-boom-bust-data-center-ai-trump/801991/

Additional References

24. Source: itpro.com
Link:https://www.itpro.com/infrastructure/data-centres/data-center-industry-faces-ticking-power-time-bomb

25. Source: linkedin.com
Link:https://www.linkedin.com/posts/dr-irene-di-martino-88481823_datacenters-ai-energyinfrastructure-activity-7425556038529155072-tORi

26. Source: latitudemedia.com
Link:https://www.latitudemedia.com/news/report-global-grid-congestion-puts-20-of-data-center-projects-at-risk/

Published: April 10, 2025

27. Source: GOV.UK
Link:https://www.gov.uk/government/consultations/accelerating-electricity-network-connections-for-strategic-demand/accelerating-electricity-network-connections-for-strategic-demand-accessible-webpage

28. Source: businessinsider.com
Title: Business Insider Trump wants Big Tech to build its own power plants
Link:https://www.businessinsider.com/data-centers-on-site-power-plants-trump-announces

29. Source: theguardian.com
Link:https://www.theguardian.com/technology/2026/may/13/datacentres-electricity-consumption-uk-us-ai

30. Source: energy.ec.europa.eu
Title: focus data centres energy hungry challenge 2025 11 17 en
Link:https://energy.ec.europa.eu/news/focus-data-centres-energy-hungry-challenge-2025-11-17_en

31. Source: brookings.edu
Title: global energy demands within the ai regulatory landscape
Link:https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/

32. Source: troutman.com
Link:https://www.troutman.com/wp-content/uploads/2026/03/Troutman_Tamarindo_Off-Grid-Data-Centers_final.pdf

33. Source: enkiai.com
Title: gas to power boom ai drives 2026 on site energy shift
Link:https://enkiai.com/data-center/gas-to-power-boom-ai-drives-2026-on-site-energy-shift/