AI Datacenter Buildout: Cancellation Myths, Geographic Shifts, and Policy Enablement

Three related debates are running simultaneously about US AI datacenter construction. First, a widely circulated claim that roughly half of 2026 US datacenter capacity has been canceled or delayed is being contested as methodologically flawed [4]. Second, Northern Virginia — the dominant US datacenter market for two decades — is losing market share faster than analysts expected [5]. Third, FERC's new large-load interconnection framework is enabling AI facilities to self-fund grid upgrades and clear interconnection queues in as little as 60 days by committing to flexible load, with NVIDIA already building facilities designed as flexible grid assets [6].

The cancellation narrative, if accepted uncritically, misrepresents the pace of AI infrastructure buildout and can distort investment and policy decisions. The FERC ruling and geographic redistribution together suggest that where and how datacenters connect to the grid is changing more than whether they are being built at all.

A statistic claiming that roughly half of US datacenter capacity planned for 2026 has been canceled or delayed spread widely in mid-June 2026, picked up by outlets including TechSpot, The Register, and Yahoo Finance [1][2][3]. SemiAnalysis published a direct rebuttal arguing the figure originates from Sightline Climate data that systematically undercounts the actual pipeline [4]. Sightline tracks only large publicly announced projects — the subset most prone to slippage — while ignoring the hyperscaler self-build capacity that dominates actual construction. SemiAnalysis says their own bottom-up model, used by major investors and technology companies, showed only about a 1% change in North America hyperscaler self-build forecasts over the past six months, with co-location moving less than 5%. They note that the top two hyperscalers' self-build capacity alone exceeds Sightline's entire US 'under construction' estimate. SemiAnalysis also identifies AI-generated analysis as a compounding problem: models built with tools like Claude Code treat press releases and gigawatt-scale announcements as ground truth, producing reports that amplify the overcounting error.

At the same time, the geographic composition of US datacenter capacity is shifting. Northern Virginia has concentrated data center infrastructure for roughly twenty years and remains the largest US market, but is now losing market share at a pace faster than most forecasters predicted [5]. The specific drivers — power availability, land costs, regulatory environment, FERC interconnection rules — are not fully detailed in available reporting, but the trend appears established.

FERC's new large-load interconnection framework is likely one structural factor reshaping geography. The rule allows facilities like AI factories to self-fund necessary grid upgrades and, by committing to operate as flexible loads that respond to grid conditions, move through the interconnection queue in as little as 60 days rather than years [6]. NVIDIA's Vladimir Troy frames this as simultaneously pro-growth and pro-affordability, citing Lawrence Berkeley National Laboratory data showing a roughly 6-cent-per-kWh reduction in retail electricity prices for every 10% increase in state electricity consumption. NVIDIA is not waiting for market response: through its partnership with Emerald AI, it is already building AI factories designed from the ground up as flexible grid assets, with commercial deployment planned for later in 2026.

These threads intersect. FERC's framework makes new locations economically viable by solving the interconnection bottleneck that historically favored established markets. That same bottleneck has constrained Northern Virginia growth, making alternatives more attractive. And the sites most likely to appear in cancellation statistics — large publicly announced projects in congested markets — may be the very projects facing the most friction, while hyperscaler self-builds and FERC-enabled flexible-load facilities in new geographies proceed on schedule.