Infrastructure & Buildout — Artificial Minds

U.S. Infrastructure Landscape

28 selected projects

Project Map — United States

Hover for details · click to expand in table

Capacity

Status

End User

Key Findings — March 2026

AI infrastructure announcements continue to accelerate, but the underlying delivery pipeline remains constrained. Four signals matter most for underwriting discipline today.

1. Energized capacity remains a small fraction of announced supply

Only 4.1 GW of the capacity across the 28 projects tracked is operational today, while announced capacity totals several tens of gigawatts. The gap between announced projects and energized capacity remains the central underwriting risk. Delivery timelines are determined less by capital availability and more by power access, permitting, and construction sequencing.

2. Tenant visibility remains limited outside hyperscaler-owned campuses

14 of the 28 projects have a confirmed end user. Most of these are hyperscaler-owned or hyperscaler-anchored developments. Independent developers frequently announce multi-gigawatt campuses without disclosing a tenant. At present, TECfusions' TensorWave lease is one of the few disclosed anchors in an independent AI campus. Tenant uncertainty remains a primary financing constraint.

3. A significant share of projects remain at announcement stage

Eight of the tracked projects have neither permits in hand nor a confirmed anchor tenant. These developments exist primarily as site announcements rather than executable infrastructure plans. Projects at this stage rarely build at the scale or timeline initially described, particularly when grid upgrades or power generation must be secured.

4. Transmission constraints are concentrating risk in key AI corridors

Central Ohio illustrates the emerging bottleneck. Meta Prometheus and Google New Albany are located in the same AEP transmission corridor. AEP has indicated that large transmission upgrades in this region may require seven to ten years. Even projects with confirmed financing and construction underway face delivery risk when transmission expansion becomes the gating constraint.

5. Behind-the-meter generation accelerates development but introduces new dependencies

Many independent projects bypass interconnection queues through behind-the-meter gas generation. This approach reduces grid timelines but introduces different constraints. Turbine manufacturing lead times remain 18 to 24 months, air permitting can delay deployment, and projects become dependent on reliable gas supply infrastructure.

6. Hyperscaler build timelines are resetting market expectations

Recent projects have demonstrated unusually rapid deployment timelines. xAI's Colossus cluster moved from construction to operation in under twelve months, and other facilities have been converted from existing industrial structures in weeks. These builds relied on atypical conditions but have materially shifted expectations around delivery speed across the ecosystem.

7. Nuclear power is becoming part of long-run hyperscaler energy strategy

Major hyperscalers have begun signing long-duration nuclear supply agreements. Microsoft has partnered with Constellation to restart Three Mile Island. Amazon has contracted with Talen Energy. Meta has announced agreements tied to emerging nuclear developers. These contracts indicate that hyperscalers are seeking firm, carbon-free capacity for long-term AI infrastructure expansion.

These signals suggest that the next phase of AI infrastructure development will be determined less by announced capital commitments and more by power delivery, permitting execution, and tenant visibility.

How to use this tracker

The AI Infrastructure Buildout Tracker maps 28 major AI data center projects across the United States. It tracks where infrastructure is operational, under construction, or still announced, along with developer, location, power capacity, and capex where available. Clicking a project expands a short project brief with overview, power strategy, and key risks. Some projects include a full deep dive with detailed research on infrastructure, power systems, contractors, and execution risks. The goal is to track the real delivery pipeline of AI infrastructure, not just announcements.

AI INFRASTRUCTURE BUILDOUT TRACKER

Loading…

Project	Developer	State	Status	Planned Capacity	CAPEX	Capital Certainty

※

CAPEX Disclosure: CAPEX figures are shown only where publicly disclosed by the developer or tenant. Most AI infrastructure projects do not publish full project cost due to phased development and hardware procurement variability.

What is being built,and how much of it is real?

Markets & Capital

Policy & Regulation

Power & Utilities

Sustainability & Community

Global Dynamics & Geopolitics

Explore All Research Tracks

What is being built,
and how much of it is real?