The grid bottleneck behind AI abundance

Introduction

One of the least visible constraints on an AI-powered future is not a shortage of energy in the abstract. It is the difficulty of getting large amounts of electricity delivered to the exact place where new computing facilities want to operate. A company can secure power purchase agreements, finance a new data-centre campus and order advanced chips, yet still wait years for a grid connection.

That matters because the most ambitious visions of AI abundance depend on enormous growth in computing capacity. If advanced AI is to accelerate medicine, scientific discovery, education, robotics and climate technologies, then the physical infrastructure behind computation must expand as well. Increasingly, the obstacle is not whether enough electricity exists somewhere in the wider system. It is whether local grids, substations, transformers and transmission lines can absorb new loads quickly enough. In many regions, connection queues, equipment shortages and network upgrades are becoming a hidden bottleneck that shapes where AI can grow and how fast.[IEA]iea.orgExecutive summary – Key Questions on Energy and AI14 Apr 2026 — Constrained by slow grid connections, data centre developers in the Un…

Why location matters more than global electricity totals

Public discussion often treats electricity as a single global pool. In reality, power systems are geographical networks. Electricity has to be generated, transmitted and distributed through specific infrastructure. A data centre does not need power “somewhere”; it needs reliable power at its own site, every hour of the day.

This distinction helps explain why arguments about total electricity supply can miss the real constraint. A country may have enough generation capacity on paper while still lacking the local transmission infrastructure needed to serve a new gigawatt-scale AI campus. The problem resembles a city with enough water overall but insufficient pipes to supply a fast-growing district.

The International Energy Agency has warned that slow grid connections are becoming a major constraint on data-centre development. In the United States, some developers have responded by pursuing projects with dedicated on-site gas generation because conventional grid connections can take so long.[IEA]iea.orgData centre electricity use surged in 2025, even with…Apr 16, 2026 — New IEA report explores AI's growing energy footprint, options fo…

This is one reason AI infrastructure is clustering in a relatively small number of locations. Northern Virginia’s “Data Center Alley”, parts of Texas, the Nordic countries and selected industrial regions possess combinations of transmission capacity, utility support, land availability and cooling advantages that are difficult to replicate quickly elsewhere. As demand accelerates, those advantages become increasingly valuable. RMI[World Economic Forum]weforum.orgWorld Economic ForumIs power grid connectivity the strategic bottleneck for AI?May 18, 2026 — Investment in AI data centres is growing fa…Published: May 18, 2026

The result is that the relevant question for AI abundance is often not “Can civilisation generate enough electricity?” but “Can the right amount of electricity reach the right place at the right time?”

How substations, transformers and queues slow expansion

Grid connection delays are rarely caused by a single failure. More often they emerge from a chain of physical and administrative constraints.

When a hyperscale AI facility requests a connection, utilities and grid operators must determine whether local infrastructure can support the new load. If not, upgrades may be required. Those upgrades can include:

Each step can trigger further delays.

A modern AI campus can require hundreds of megawatts of power and, in some proposals, more than a gigawatt. That is comparable to the electricity demand of a large city. Existing infrastructure was often not designed with multiple city-scale loads suddenly appearing in a concentrated area.[World Economic Forum]weforum.orgWorld Economic ForumIs power grid connectivity the strategic bottleneck for AI?May 18, 2026 — Investment in AI data centres is growing fa…Published: May 18, 2026

Transformer shortages have become especially important. Large power transformers are specialised industrial products with long manufacturing lead times. Reports from the data-centre industry describe lead times stretching to several years, creating situations where facilities are physically ready but cannot obtain critical electrical equipment quickly enough.[Tom's Hardware]tomshardware.comdata center developments are expected to be delayed or canceled due to severe shortages in power infrastructure components, despite over…

Connection queues create another layer of delay. Grid operators must evaluate large numbers of proposed projects and determine how each affects system reliability. Historically, many interconnection systems were designed around slower growth patterns. They were not built for a surge of simultaneous requests from renewable-energy developers, battery projects and giant AI campuses.

In parts of the United States, interconnection timelines have lengthened dramatically. Analyses of grid-connection processes show average waits that have grown from less than two years in the late 2000s to many years today. PJM, the regional transmission organisation covering much of the Mid-Atlantic and Midwest, has become a particularly important example because it serves regions experiencing heavy data-centre growth.[Yahoo Finance]finance.yahoo.comFinance AI data centers have U.Spower grid struggling to keep up3 days ago — The average time from interconnection application to commercial operation has risen from les…

The consequence is a mismatch between software timelines and infrastructure timelines. AI models can improve in months. Transmission projects may require a decade.

The hidden competition between AI and everything else

The bottleneck is not solely about AI.

The same grid infrastructure is also being asked to support electrified transport, new manufacturing facilities, heat pumps, battery projects and clean-energy generation. A transformer used for an AI campus cannot simultaneously be installed elsewhere. Transmission capacity allocated to one major load is unavailable for another.

This creates a subtle political challenge for the AI bloom vision. The optimistic case for advanced AI often assumes rapid growth in both computation and clean energy. Yet the same queues that slow AI facilities can also slow renewable-energy projects attempting to connect to the grid. Clean Power Association materials note that generator interconnection processes can already take years before projects reach operation.[ACP]cleanpower.orgACPInterconnection 101 – Fact SheetInterconnection is the complex process of connecting new electricity generators – like wind, solar, an…

As a result, AI growth does not occur against a static background. It enters an infrastructure system already under strain.

In Denmark, grid operators recently paused new large-scale connection requests after a surge in demand. Reports described roughly 60 gigawatts of requested connections, including major data-centre projects. The episode illustrated how quickly demand projections can outrun planning assumptions even in countries with strong clean-energy resources.[Tom's Hardware]tomshardware.comdata center developments are expected to be delayed or canceled due to severe shortages in power infrastructure components, despite over…

This does not mean AI and decarbonisation are inherently in conflict. AI could help design better grids, improve forecasting and accelerate clean-energy deployment. But in the near term they can compete for scarce physical infrastructure, creating difficult choices about priorities and cost allocation.

Why delays matter for the long-term AI bloom case

At first glance, a three-year or five-year connection delay may seem like a routine infrastructure problem. In the context of AI, however, timing matters unusually much.

Many arguments about AI-driven abundance depend on rapid compounding. If more capable AI systems help accelerate research, automate engineering and improve economic productivity, then delays in compute deployment can slow the entire feedback loop. The question is not merely whether additional data centres are eventually built. It is whether they arrive soon enough to participate in successive generations of capability growth.

This is one reason grid access has become strategically important. Companies increasingly compete not only for chips and talent but also for locations with available electrical capacity. Industry discussions increasingly frame access to power as a key determinant of future compute supply.[World Economic Forum]weforum.orgWorld Economic ForumIs power grid connectivity the strategic bottleneck for AI?May 18, 2026 — Investment in AI data centres is growing fa…Published: May 18, 2026

The issue becomes even sharper if future AI systems require large-scale inference rather than occasional training runs. Training a frontier model is an enormous event, but serving billions of users, autonomous agents or scientific systems requires continuous computing infrastructure. That shifts attention from isolated energy purchases towards long-term grid integration.

In this sense, grid connections can function as a hidden governor on the speed of AI expansion. Software may scale exponentially while electrical infrastructure expands through permitting processes, construction schedules and industrial supply chains.

Why some companies are moving power to the data centre

Because grid timelines are so slow, some developers are experimenting with alternative approaches.

One strategy is to colocate data centres with dedicated generation assets. Another is to build facilities near regions with surplus renewable power and available transmission capacity. Others are exploring campuses that combine generation, storage and computing within a single integrated project.[Hanwha Data Centers]hanwhadatacenters.comthe data center power chain from grid to rackdata center grid power demand is projected to nearly triple by 2030, reaching over 134 GW, while interconnection queues now average five…

There is also growing interest in making AI workloads more flexible. Traditional planning often treats data centres as fixed loads that require constant power regardless of grid conditions. Researchers are increasingly examining whether some forms of AI computation can be shifted across time or geography, reducing pressure on constrained networks. Early modelling suggests that workload flexibility could lower some grid-investment requirements and reduce congestion costs, though the benefits depend heavily on location and operating patterns.[arXiv]arxiv.orgTo Defer or To Shift? The Role of AI Data Center Flexibility on Grid InterconnectionApril 7, 2026…Published: April 7, 2026

Storage systems represent another possible pressure valve. Batteries can help smooth sudden demand spikes, provide backup power and support grid stability. Research on grid-integrated AI facilities suggests that large-scale storage could allow data centres to become more active participants in electricity systems rather than purely passive consumers.[arXiv]arxiv.orgTo Defer or To Shift? The Role of AI Data Center Flexibility on Grid InterconnectionApril 7, 2026…Published: April 7, 2026

None of these approaches fully eliminates the need for stronger grids. They are better understood as ways to reduce stress on existing infrastructure while broader expansion proceeds.

What better grid planning would change for the public

Grid connection delays may sound like a technical issue relevant only to utilities and technology companies. In practice, they shape whether the benefits of advanced AI become widely available.

If grid expansion lags far behind compute demand, several outcomes become more likely:

By contrast, faster and better-coordinated grid planning could make AI growth more compatible with broader public goals.

That means building transmission before shortages become acute rather than after. It means modernising interconnection procedures that were designed for slower growth eras. It means expanding transformer manufacturing and other electrical supply chains. It also means deciding openly who should pay for upgrades: households, taxpayers, utilities, technology companies or some combination of the four.[Broadband Breakfast]broadbandbreakfast.comBroadband BreakfastPermitting Delays Could Drive Up PJM Power Costs…4 days ago — WASHINGTON, May 13, 2026 – Delays in permitting and g…Published: May 13, 2026[The Washington Post]washingtonpost.comgovernors and the White House announced a plan to address soaring electricity costs driven by expanding data center demand, particularly…

The most optimistic AI abundance scenario assumes that intelligence becomes dramatically cheaper and more available. Yet intelligence alone does not move electrons through a transformer. The long-term future imagined by AI bloom depends not only on breakthroughs in algorithms but also on the unglamorous infrastructure that allows those algorithms to run at civilisation scale.

In that sense, one of the decisive questions for an AI-rich future may be surprisingly mundane: not whether humanity can invent more intelligence, but whether it can connect the next substation quickly enough.

Amazon book picks

Further Reading

Books and field guides related to The grid bottleneck behind AI abundance. 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

Make It Stick

By Peter C. Brown, Henry L. Roediger III et al.

Directly addresses whether learners can perform later without support.

See on Amazon

Book

How Learning Happens

By Paul A. Kirschner, Carl Hendrick

Explains transfer, practice and memory mechanisms.

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

Browse more on Amazon:Human CompatibleMake It StickHow Learning Happens

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

'Electricity Pylon' Wall Art Poster Prints (PP056342)

SearcheBay.co.uk: electrical grid poster

Browse similar oneBay.co.uk

Example eBay listing

Vtg DC-10 Full Color Cockpit Flight Engineers Panel Diagram 1983 Series 30

SearcheBay.co.uk: electrical grid poster

Browse similar oneBay.co.uk

Example eBay listing

'Electric Pylon' Wall Art Poster Prints (PP046515)

SearcheBay.co.uk: electrical grid poster

Browse similar oneBay.co.uk

Example eBay listing

'Power Transmission Structures' Wall Art Poster Prints (PP056389)

SearcheBay.co.uk: electrical grid poster

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/reports/key-questions-on-energy-and-ai/executive-summary

2. 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

3. Source: rmi.org
Title: pjms speed to power problem and how to fix it
Link:https://rmi.org/pjms-speed-to-power-problem-and-how-to-fix-it/

4. Source: insidelines.pjm.com
Title: board outlines plans to integrate large loads reliably
Link:https://insidelines.pjm.com/pjm-board-outlines-plans-to-integrate-large-loads-reliably/

5. Source: finance.yahoo.com
Title: Finance AI data centers have U.S
Link:https://finance.yahoo.com/sectors/energy/articles/ai-data-centers-u-power-165930006.html

6. Source: cleanpower.org
Link:https://cleanpower.org/resources/interconnection/

7. Source: arxiv.org
Link:https://arxiv.org/abs/2604.05376

Published: April 7, 2026

8. Source: arxiv.org
Link:https://arxiv.org/abs/2605.14109

9. Source: arxiv.org
Link:https://arxiv.org/abs/2602.01013

Published: February 1, 2026

10. Source: arxiv.org
Link:https://arxiv.org/abs/2603.00415

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

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

13. Source: iea.org
Link:https://www.iea.org/reports/energy-and-ai/ai-and-energy-security

14. Source: build.inc
Title: pjm load growth data center underwriting
Link:https://build.inc/insights/pjm-load-growth-data-center-underwriting

15. Source: pjm.com
Title: 20250509 item 02 large load additions workshop presentation
Link:https://www.pjm.com/-/media/DotCom/committees-groups/workshops/llaw/2025/20250509/20250509-item-02—large-load-additions-workshop—presentation.pdf

Published: May 2025

16. Source: arxiv.org
Link:https://arxiv.org/html/2509.07218v3

17. Source: rmi.org
Title: interconnection reform ai data centers generator queues
Link:https://rmi.org/interconnection-reform-ai-data-centers-generator-queues/

18. Source: energy.gov
Link:https://www.energy.gov/oe/clean-energy-resources-meet-data-center-electricity-demand

19. Source: weforum.org
Link:https://www.weforum.org/stories/2026/05/electricity-data-grid-connectivity-strategic-bottleneck-ai-transformation/

Published: May 18, 2026

20. Source: tomshardware.com
Link:https://www.tomshardware.com/tech-industry/artificial-intelligence/half-of-planned-us-data-center-builds-have-been-delayed-or-canceled-growth-limited-by-shortages-of-power-infrastructure-and-parts-from-china-the-ai-build-out-flips-the-breakers

21. Source: tomshardware.com
Title: Tom’s Hardware Denmark presses pause on new data center grid
Link:https://www.tomshardware.com/tech-industry/artificial-intelligence/denmark-presses-pause-on-new-data-center-grid-connections-as-total-requests-hit-60-gw-nordic-nation-is-the-latest-to-put-the-brakes-on-ai-buildouts

Published: May 6, 2026

22. Source: hanwhadatacenters.com
Title: the data center power chain from grid to rack
Link:https://www.hanwhadatacenters.com/blog/the-data-center-power-chain-from-grid-to-rack/

23. Source: broadbandbreakfast.com
Link:https://broadbandbreakfast.com/permitting-delays-could-drive-up-pjm-power-costs-by-100-billion/

Published: May 13, 2026

24. Source: washingtonpost.com
Link:https://www.washingtonpost.com/business/2026/01/16/pjm-data-centers-electricity-bipartisan/

25. Source: Wikipedia
Link:https://en.wikipedia.org/wiki/Interconnection

26. Source: hanwhadatacenters.com
Link:https://www.hanwhadatacenters.com/blog/top-energy-companies-for-ai-data-centers-2025-power-guide/

27. Source: hanwhadatacenters.com
Title: Learn why electric utilities and grid bottlenecks delay
Link:https://www.hanwhadatacenters.com/blog/data-center-grid-limitations-the-power-bottleneck/

Additional References

28. Source: reuters.com
Link:https://www.reuters.com/business/energy/eon-unit-signs-200-mw-grid-connection-atnorth-ai-data-centre-sweden-2026-05-28/

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

30. Source: merriam-webster.com
Link:https://www.merriam-webster.com/dictionary/interconnection

31. Source: techradar.com
Link:https://www.techradar.com/pro/inference-requires-infrastructure-that-is-closer-to-users-and-more-energy-efficient-antimatter-debuts-global-ai-network-built-to-bypass-grid-bottleneck-with-a-400-000-gpu-roll-out-by-2030-across-1000-data-centers

32. Source: linkedin.com
Link:https://www.linkedin.com/posts/craig-sullivan-a989136_ai-data-center-projects-are-facing-significant-activity-7454351469006311424-VsJ6

33. Source: linkedin.com
Link:https://www.linkedin.com/posts/international-energy-agency_data-centre-electricity-use-surged-in-2025-activity-7450847869810536448-GGzW

34. Source: datacenterknowledge.com
Link:https://www.datacenterknowledge.com/build-design/ai-data-center-boom-rewires-us-power-supply-chain

35. Source: linkedin.com
Link:https://www.linkedin.com/posts/magdyaly_grid-interconnection-bottleneckred-flags-activity-7398571071349620736-gUfT

36. 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

37. Source: renewableenergyworld.com
Title: pjm is getting serious about reliably serving large load customers
Link:https://www.renewableenergyworld.com/power-grid/outage-management/pjm-is-getting-serious-about-reliably-serving-large-load-customers/