Stop Saying Half of 2026 US Datacenter Capacity Is Canceled:
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-06-18
SemiAnalysis researchers refute the widely-circulated claim that half of 2026 US datacenter capacity is canceled, presenting bottom-up filing analysis showing North American hyperscaler forecasts moved less than 1% over six months and attributing the narrative to AI-generated estimates built on flawed, announcement-based data.
Extraction
Topics: datacenter-capacityai-infrastructuremarket-analysissupply-chain
Claims
- SemiAnalysis's YE2026 North America hyperscaler self-build forecast changed by only ~1% over the past six months, with colocation moving less than 5%.
- The widely-cited '50% delayed' figure originates from Sightline Climate data that undercounts construction capacity by multiples, tracking only large publicly announced projects most prone to slippage.
- AI-generated datacenter models built with Claude Code treat press releases and GW-scale announcements as ground truth, producing systematically inaccurate overestimates of delays.
- Taking only the top two hyperscalers' self-build capacity alone exceeds Sightline's entire 'under construction' estimate for the US market.
- Real, specific delays are occurring—such as STACK Infrastructure/Oracle's site being pushed to 2029—but these do not represent a systemic half-of-capacity collapse.
Key quotes
over the last 6 months, our YE2026 NA Hyperscaler Self-build forecast only moved by ~1%, and NA colocation <5%.
Claude Code pulls press releases, views unfounded GW-scale announcements as ground truth, misunderstands construction timelines and grid complexities, and compiles a terribly inaccurate report.
The '50% cancelled or delayed' figure isn't really a statement about the US datacenter pipeline, but rather about the slice of the pipeline most prone to slipping.