📊 Full opportunity report: The queue. Why the grid, not the chip, is the binding constraint on AI. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

TL;DR

The primary bottleneck for AI infrastructure expansion has shifted from chip supply to the US power grid interconnection queue. Capital is bypassing the grid, leading to private power solutions and political disputes over who bears the costs.

The US power grid interconnection queue has become the primary bottleneck for AI infrastructure expansion, overtaking chip supply constraints. This shift is prompting industry players to build private power sources to bypass the grid, while raising political debates over who bears the costs of upgrading the shared infrastructure.

For the past two years, the dominant narrative centered on shortages of GPUs and chip manufacturing capacity. However, recent data shows that the real constraint now lies in the interconnection process—specifically, the long wait times for connecting new power generation projects to the grid. Currently, between 2,300 and 2,600 gigawatts of capacity are stuck in US interconnection queues, with median wait times approaching five years, and some projects facing delays up to twelve years, according to industry sources.

Demand for power from data centers and AI-related infrastructure is surging. US data-center power demand is projected to reach approximately 76 gigawatts in 2026, up from about 50 gigawatts in 2024, with global consumption potentially exceeding 1,000 terawatt-hours annually by the early 2030s. In Texas, interconnection requests for large loads increased by 700% within a single year, illustrating the scale of demand. Meanwhile, utilities like PJM report that the cost of connecting new data centers has skyrocketed, with transmission costs passing onto ratepayers, fueling political controversy.

As a response, capital is increasingly routing around the grid. Private power generation—such as behind-the-meter gas plants, co-located nuclear reactors, and onsite renewable projects—is being built to meet immediate needs. Notably, Microsoft has partnered to restart the Three Mile Island nuclear reactor to secure baseload power. These private solutions allow companies to bypass lengthy interconnection queues, but shift costs onto ratepayers and the broader grid system, raising questions about fairness and the future of infrastructure development.

The Queue — Thorsten Meyer AI
QUEUE
● DISPATCH / MAY 2026
THORSTEN MEYER AI · AI ENERGY & INFRASTRUCTURE · § 02
AI ENERGY · 02
INTERCONNECTION / QUEUE
Essay · Energy-Infrastructure Structural Reading · 2026-05-23

The queue.Why the grid, not the chip,
is the binding constraint on AI.

2,300 gigawatts are stuck in line — more than the country’s entire installed power capacity. So capital builds around the line.
For two years the AI buildout was a chip story. That story is over. The binding constraint is the grid — and the line you wait in to connect to it. Roughly 2,300-2,600 GW of capacity is stuck in US interconnection queues, more than the entire installed fleet; the median wait approaches five years, some data centers face twelve, and ~80% of projects withdraw. The demand hitting that queue: US data-center power ~76 GW by 2026, CenterPoint’s large-load requests up 700% in a year. So capital routes around it — a behind-the-meter gas plant builds in ~18 months vs grid access maybe 2035; Microsoft restarted Three Mile Island for 835 MW of baseload, bypassing transmission. But the bypass has a cost it does not bear: $1.98B of transmission cost landed on Virginia ratepayers; PJM’s capacity auction ran $2.2B → $14.7B. The structural argument: the grid is the bottleneck, and the response is a parallel private grid that solves time-to-power for whoever has the capital — and externalizes the cost of the shared grid onto everyone else.
2,300 GW
Stuck in US interconnection queues
more than total installed capacity
~5 yr
Median wait to commercial operation
up to 12 years for data centers
~18 mo
Behind-the-meter gas build time
vs grid access maybe 2035
$1.98B
Transmission cost on Virginia
ratepayers · the cost-shift, concrete
THE QUEUE· THE GRID IS THE BINDING CONSTRAINT· 2,300-2,600 GW STUCK· MORE THAN TOTAL INSTALLED CAPACITY· ~5-YEAR MEDIAN WAIT · UP TO 12· ~80% OF PROJECTS WITHDRAW· US DATA-CENTER ~76 GW BY 2026· CENTERPOINT +700% IN A YEAR· BTM GAS ~18 MONTHS· THREE MILE ISLAND RESTART · 835 MW· POWER-CERTAIN SITES +15-25% LEASE· PJM AUCTION $2.2B → $14.7B· VIRGINIA RATEPAYERS $1.98B· RATEPAYER PROTECTION PLEDGE· MICROSOFT 40 GW CONTRACTED· CHINA +430 GW/YEAR· THE SEARCH FOR MEGAWATTS· A BIFURCATED BUILDOUT· THE QUEUE· THE GRID IS THE BINDING CONSTRAINT· 2,300-2,600 GW STUCK· MORE THAN TOTAL INSTALLED CAPACITY· ~5-YEAR MEDIAN WAIT · UP TO 12· ~80% OF PROJECTS WITHDRAW· US DATA-CENTER ~76 GW BY 2026· CENTERPOINT +700% IN A YEAR· BTM GAS ~18 MONTHS· THREE MILE ISLAND RESTART · 835 MW· POWER-CERTAIN SITES +15-25% LEASE· PJM AUCTION $2.2B → $14.7B· VIRGINIA RATEPAYERS $1.98B· RATEPAYER PROTECTION PLEDGE· MICROSOFT 40 GW CONTRACTED· CHINA +430 GW/YEAR· THE SEARCH FOR MEGAWATTS· A BIFURCATED BUILDOUT·
FIG. 01 — THE BINDING CONSTRAINT MOVED
From the chip you manufacture to the grid you wait in line for
When site selection is driven by where you can get power, the binding constraint has moved
2021-2024 · The chip era
Compute
GPU allocation, fab capacity, export controls. Partnerships around cloud, hardware supply, software. The assumption: chips + capital = data center.
2025-2026 · The grid era
Power
Megawatts, queue position, transmission, time-to-power. Partnerships around energy. The search for megawatts now beats latency and fiber in site selection.
Chips can be manufactured faster than grids can be expanded, which is why the constraint moved to the grid the moment chip supply loosened. The data center can be designed, financed, and built in 18-24 months. The grid connection it needs can take five to twelve years. That maturity gap — between the rapid innovation cycle of data-center technology and the slow, linear deployment of grid infrastructure — is the single greatest constraint on the buildout.
FIG. 02 — ANATOMY OF THE QUEUE · WHY IT TAKES FIVE YEARS
Four compounding bottlenecks on a process built for a slower era
FERC Order 2023 fixes the easiest one — the study backlog — while the harder ones increasingly dominate
01
Utility study backlogs
Request volume far outpaces what utilities have ever processed; studies are sequential and under-resourced.
02
Transmission upgrades
New substations, lines, reconductoring — years to build, and the cost is contested.
03
Permitting complexity
Multiple jurisdictions, each with its own timeline and veto points; increasingly the binding step.
04
Equipment lead times
High-voltage transformers now carry multi-year lead times. Even an approved project waits for hardware.
Nearly 80% of projects in the queue eventually withdraw — speculative projects occupying study slots and slowing the viable ones behind them. LBNL: interconnection wait times have more than doubled in 15 years. FERC Order 2023’s “first-ready, first-served” cluster model addresses the study backlog — but the harder bottlenecks (transmission, permitting, transformers) are the ones increasingly dominating. The queue is not congestion that clears; it is a structural mismatch between the speed of demand and the speed of connection.
FIG. 03 — THE DEMAND WALL · WHAT IS HITTING THE QUEUE
A step-change in scale, density, and utilization the grid was not designed for
A single data-center campus can now request more power than a utility’s historical peak demand
2024 · US data-center demand
~50 GW
2026 · US data-center demand
~76 GW
by 2030 · added capacity needed
>150 GW
Global data-center consumption could exceed 1,000 TWh annually by the early 2030s (up from 460 TWh in 2022). Hyperscale (100+ MW) is ~41% of worldwide capacity; single campuses of 1 GW+ — a large nuclear unit’s output — are now explored by single developers. The utility shock: CenterPoint’s large-load requests grew 700% in a year (1→8 GW), and ComEd, PPL, and Oncor report more GWs of data-center applications than their historical maximum peak demand. Data centers run near 100% utilization — constant baseload, not peaky load served from reserve margin.
FIG. 04 — ROUTING AROUND THE QUEUE · THE BYPASS
Every form of the bypass is a way to get power without waiting in line
Available to whoever has the capital to self-generate — which is the seam
BYPASS
HOW IT WORKS
TIME-TO-POWER
Behind-the-meter gas
On-site generation behind the utility meter · midstream gas pivots to on-site power provider · Foley 2026: 56% of developers exploring
~18 movs grid ~2035
Nuclear co-location
Tie directly to operating/restarting reactor, bypass transmission · Three Mile Island Unit 1 restart, 835 MW baseload
+15-25%lease premium
Flexible / interruptible
Draw from grid only when spare capacity exists · Nvidia-backed Emerald AI, 96 MW Manassas VA
Connectswhere firm can’t
Stranded-power hunt
Hunt unallocated capacity; diversify to under-utilized grids · Idaho, Louisiana, Oklahoma over Northern Virginia
Geographyrepriced
The common thread is time-to-power: an 18-month private plant or a nuclear co-location beats a decade-long queue, and the best-capitalized players are choosing to build their own power. Microsoft has surpassed Amazon as the world’s largest clean-power buyer — ~40 GW contracted — and the big four accounted for roughly half of all global clean-energy PPAs in 2025. The bypass is rational, fast, and available only to those with the capital to self-generate.
FIG. 05 — WHO PAYS FOR THE BYPASS · THE COST-SHIFT
The bypass solves the developer’s problem and relocates the grid’s cost onto ratepayers
The benefit accrues to the data center; the cost of the grid it depends on is socialized
$2.2→14.7B
PJM capacity auction
in a single year
$1.98B
Transmission cost on
Virginia ratepayers (2024)
~$7B
More in higher rates
across PJM consumers
Virginia’s residents are paying nearly $2 billion to connect data centers they do not own and whose power they do not consume.
When a data center self-generates behind the meter but still relies on the grid for backup, it avoids much of the cost while retaining the benefit — the bypass at its most extractive. The early-March 2026 White House Ratepayer Protection Pledge is nonbinding, and covers generation, not the larger transmission-and-capacity burden. The politics of AI energy is not about whether to build — it is about who pays for the grid the buildout requires. The default, absent regulation, is “everyone, whether or not they benefit.”
The grid is the bottleneck. The private grid is the response. And the seam between them — who pays for the public infrastructure the private builders still lean on — is where the economics and politics of the AI buildout are now decided.
Thorsten Meyer · The Queue · AI Energy & Infrastructure 02

Implications of the Grid Constraint for AI Infrastructure

The shift of the bottleneck from chip manufacturing to the power grid fundamentally alters the landscape of AI infrastructure development. It accelerates a bifurcation: well-capitalized firms can build private power sources to bypass the grid delay, while others remain dependent on a congested, slow-moving public system. This dynamic re-prices geography, as proximity to reliable power becomes more critical than latency or fiber infrastructure. It also shifts the cost burden onto ratepayers, fueling political conflicts and potentially deepening infrastructure inequality. Ultimately, the grid constraint and the private solutions it spurs could shape the geographic distribution, costs, and political landscape of AI buildout for years to come.

Amazon

private renewable energy generators for data centers

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

From Chip Shortages to Power Grid Bottlenecks

Initially, the focus of AI infrastructure constraints centered on semiconductor supply chains, with global chip shortages limiting deployment. Over time, it became clear that the bottleneck was not just hardware but also the infrastructure needed to power AI systems. The US faced a growing backlog in grid interconnection requests, with thousands of gigawatts of projects waiting years for connection approval. This backlog has been compounded by rising demand, especially from data centers and hyperscalers, which are seeking reliable, large-scale power sources to support AI growth. China, by contrast, continues to add hundreds of gigawatts of capacity annually, highlighting the US’s unique interconnection challenge.

“The interconnection queue is now the binding constraint on AI infrastructure growth, shifting the focus from silicon shortages to grid access.”

— Thorsten Meyer

XLUX Long Probe Deep Use Soil Moisture Meter, Water Monitor Indicator Sensor, Hygrometer for Outdoor Indoor Large Pot Plants, Flower, Gardening, Farming

XLUX Long Probe Deep Use Soil Moisture Meter, Water Monitor Indicator Sensor, Hygrometer for Outdoor Indoor Large Pot Plants, Flower, Gardening, Farming

Accurate Soil Moisture Detection: The XLUX Soil Moisture Meter can tell you if the soil deep inside your…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Unclear Long-Term Effects of Private Grid Bypass

It remains uncertain how widespread and sustainable private power solutions will become, and whether policy interventions will address the rising costs and delays in grid interconnection. The political battle over who pays for grid upgrades continues to evolve, with potential reforms on the horizon, but details are still emerging.

Nuclear Reactor Patent Print Nuclear Engineering Science Stainless Steel Insulated Tumbler

Nuclear Reactor Patent Print Nuclear Engineering Science Stainless Steel Insulated Tumbler

The illustration presents a detailed patent-style breakdown, showing mechanical components arranged in precise, labeled views. Its technical layout…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps in Addressing Grid Constraints and Costs

Industry and policymakers are likely to focus on accelerating interconnection processes, reforming cost allocation, and regulating private power solutions. Monitoring how these developments influence the geographic distribution of AI infrastructure and the political landscape will be critical. Additionally, technological innovations in grid modernization and energy storage could mitigate some bottlenecks, but their impact remains to be seen in the short term.

ECO-WORTHY 10KW Output 9.36KWH Off Grid Solar System Complete Kit for Home:12pcs 195W Solar Panels(2340W)+10KW 120V/240V Inverter (UL 1741)+ 2pcs 48V 100AH Lithium Batteries(UL 1973&UL 9540A)

ECO-WORTHY 10KW Output 9.36KWH Off Grid Solar System Complete Kit for Home:12pcs 195W Solar Panels(2340W)+10KW 120V/240V Inverter (UL 1741)+ 2pcs 48V 100AH Lithium Batteries(UL 1973&UL 9540A)

[Ideally Output of 9.36KWH] The power of 9.36KWh per day under 4 hours full sunshine by the 2340W…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

Why has the focus shifted from chip shortages to the power grid?

The interconnection queue delays now exceed chip supply issues in constraining AI infrastructure growth, as the bottleneck is in connecting new power capacity to the grid.

How are companies bypassing the grid constraints?

Many are building private power sources such as behind-the-meter gas plants, co-located nuclear reactors, and onsite renewables to meet immediate energy needs, avoiding long interconnection delays.

Who bears the costs of these private solutions?

The costs of bypassing the grid, including transmission and capacity investments, are often passed onto ratepayers and the broader public system, leading to political disputes.

What are the risks of relying on private power sources?

Private solutions can lead to increased inequality, politicization of infrastructure costs, and potential underinvestment in the shared grid, which could hamper future expansion and reliability.

Will policy changes address the interconnection backlog?

Policymakers are considering reforms to streamline interconnection processes and reform cost allocation, but the timeline and effectiveness of these measures remain uncertain.

Source: ThorstenMeyerAI.com

You May Also Like

Two Channels: How the Pentagon Just Split Frontier-AI Procurement in Half

The Pentagon has split its frontier AI procurement into two distinct channels, positioning Anthropic exclusively in the cybersecurity-focused stream, not the classified network.

Restartable Sequences

An in-depth look at restartable sequences (rseq), a Linux system programming feature that enhances performance on multi-core CPUs, with implications for future OS support.

732 Bytes to Root. One Hour of Scan Time.

A new Linux kernel vulnerability, Copy Fail, was discovered by Theori in just one hour of scanning, enabling root access across major distributions.

Class-action price-fixing lawsuit targets hard drive component makers as costs skyrocket — 13-year scheme allegedly drove up prices for major HDD brands

A class-action lawsuit has been filed against major manufacturers of hard drive suspension assemblies, alleging a 13-year price-fixing scheme that increased costs for consumers and resellers.