Saturday, September 5, 2026

70% of U.S. Grid Requests Withdrawn as $5.2T AI Data Center Buildout Hits Power Wall

More than 70% of U.S. grid interconnection requests are ultimately withdrawn due to capacity constraints, threatening a $5.2 trillion AI data center buildout through 2030. Goldman Sachs forecasts global data center power demand will surge 165% by 2030 versus 2023 levels. The U.S. grid was built for 1-2% annual demand growth — not an AI-driven surge of that scale.

LM Salvado

June 23, 2026

70% of U.S. Grid Requests Withdrawn as $5.2T AI Data Center Buildout Hits Power Wall
Image generated by AI for illustrative purposes. Not actual footage or photography from the reported events.
Loading stream...

More than 70% of U.S. grid interconnection requests are ultimately withdrawn, most due to capacity constraints.1 That single statistic may define the ceiling on AI infrastructure expansion through 2030.

Industry forecasts place AI data center capital expenditure at $5.2 trillion through 2030.2 Goldman Sachs projects global data center power demand will surge up to 165% by 2030 compared to 2023 levels.3 The U.S. power grid was not engineered for either number. It was built for 1-2% annual demand growth.4

A Structural Mismatch

Every new data center requires a grid interconnection agreement before it can draw power at scale. Interconnection studies take 18-24 months in many U.S. markets. Transmission upgrades require regulatory approval, environmental review, and years of construction.

When more than 70% of requests are withdrawn — often because grid capacity doesn't exist or the cost to build it is prohibitive — hyperscalers face a constraint that money alone cannot solve quickly.1

Investor Kevin O'Leary has put a blunt number on the outcome: 50% of planned U.S. data centers will never be built due to grid constraints.5 That is a forecast of cancellation, not delay.

Capex Exposed

$5.2 trillion in planned AI infrastructure investment is the headline figure.2 Grid physics could cut the actual buildout by 30-50%. The gap between announced data center projects and commissioned capacity is already widening.

Microsoft, Amazon, and Alphabet have all set aggressive expansion targets. But data center announcements and data center completions are different metrics. Interconnection queue backlogs make the distance between those two numbers measurable in years, not months.

New generation sources — nuclear, solar, gas — face the same queue. Power purchase agreements have become as competitive as land or fiber rights.

Beyond Capital

The AI infrastructure race has stopped being purely a capital allocation question. It is now a permitting, physics, and grid-policy question.

A grid designed for gradual, predictable load growth cannot rapidly absorb gigawatt-scale demand spikes. Until interconnection timelines compress or generation capacity expands materially, the $5.2 trillion buildout will run into a hard ceiling — one measured in megawatts, not dollars.13

In this story

About this analysis

This is a Via News analysis. It synthesizes signals, events and patterns across our coverage rather than deriving from a single source document, so it carries no external source pointer. Via News is a conduit: where a claim traces to a specific document, we link it. How we source

LM Salvado

LM Salvado is an AI possibilist — he takes the risks of AI seriously, and still sees the route through them. Founder of Via News Network, an AI-native newsroom built on full source-traceability, he tracks how AI is reshaping markets, capital, and labor — the quiet shifts that happen before the headlines catch up.

What we know · the intelligence behind this page
Live from the substrate
What we're seeing
AI Capital Surge Meets Investor Caution: Record Funding Rounds and Government Contracts Amid Valuation Skepticism
A single-week cluster of large AI/fintech funding rounds (Socure, Stability AI, Emerald AI, Generalist AI, Instinct, Gatik, Regent Craft) shows venture capital still pouring into AI infrastructure, identity, and autonomy plays, while Palantir's Army TITAN contract win coincided with a 6% stock drop — signaling that even flagship AI-defense revenue isn't immune to market reassessment of AI valuations. Efficiency-focused innovations like Multiverse Computing's model compression suggest the sector is also pivoting toward cost/inference economics as capital intensity draws scrutiny.
Our read on the data ›
Signals we're tracking
EPKINLY Regulatory-Clinical Success Cascade
High probability of expanded label indications, additional combination approvals, and competitive positioning strength in follicular lymphoma market. Predicts positive commercial uptake and potential accelerated review for related indications.
Patterns we're watching ›
Where sources disagree
Morgan Stanley & Co. LLC
The same metric (eps) for the same entity (Morgan Stanley & Co. LLC) reported for the identical fiscal period (Q1 2026) and observation date (2026-03-31) has two conflicting values: 3.43 USD_per_share vs 3.08 USD. This is not a temporal change — both observations claim to measure the same point in time. The ~10% discrepancy (0.35 USD difference) is material for a financial metric.
We flag conflicts openly ›
Recently verified
Checked against the original source
4,981
facts traced to their source — and we flag the ones that don't hold up.
101 entities tracked4,981 facts checked against source5,273 source documents archived
Query this data → isubstrate.com