SemiAnalysis: Half of US Data Centers to Shut Down by 2026? This Is an “AI-Coded” False Alarm
The pessimistic narrative about the slowdown in U.S. data center construction has swept the market, but the truth behind the bearish sentiment is prompting rational investors to reconsider this collective misjudgment.
On June 18, the independent industry research institute SemiAnalysis published an article explicitly refuting the extreme view widely circulating in the market that “U.S. data center construction will significantly slow down, with half of the capacity canceled or delayed by 2026.” It argues that this wave of panic stems from misreading information and sample bias, and that the actual delivered capacity of U.S. data centers in 2026 has not experienced a steep decline, with the overall construction pace of the sector remaining resilient.
Looking at the core forecast data, over the past six months, SemiAnalysis adjusted its prediction for North American hyperscale self-built data center capacity for the end of 2026 by only 1%, and colocation data center capacity fluctuations were less than 5%, with overall expectations remaining stable. The data shows that the industry’s capacity curve is far from reaching the collapse level portrayed by the market.
The institute admits that localized project delays do exist, but these are normal adjustments in large infrastructure cycles and not leading indicators of a shift in demand. Currently, pessimistic sentiment has been overly amplified by the market, and sharp volatility in sector valuations reflects more of an excessive emotional reaction than a fundamental weakening of the industry itself.
AI “blindly trusts” announcements, market “magnifies” panic
“Half of U.S. data center capacity will be canceled or delayed by 2026”—this claim has been widely spread across financial and social media recently and traces back to a Bloomberg report on April 1, which was then amplified with even more sensational headlines by several tech outlets.
But independent research institute SemiAnalysis bluntly states that this panic narrative is purely an “AI-generated false alarm”—many so-called capacity forecast models merely use AI tools to scrape press releases, treating unverified GW-scale project announcements as facts, without considering real construction timelines, grid access, and approval processes, which are critical variables.
The institute admits that while localized project delays indeed exist, the vast majority of projects marked as “canceled or delayed” are concentrated in early stages—these projects lack support from financing, permitting, or equipment orders and never actually had a realistic path to 2026 delivery.
A “mathematical error” in half delays: The issue lies in the denominator
SemiAnalysis believes that the widely cited conclusion of “50% capacity delayed or canceled in 2026” has its core issue not in the numerator—the number of canceled projects—but in a biased denominator, with sample limitations causing serious distortion of conclusions.
According to the Sightline Climate stats cited by the market, out of 12GW planned U.S. capacity for 2026, only 5GW is under construction. But SemiAnalysis independently verified using satellite imagery models that just the self-built capacity under construction from the two leading U.S. hyperscale cloud providers already exceeds 5GW, not including multi-GW projects under construction by third-party developers.
This means that Sightline’s scope only covers publicly disclosed large benchmark projects, not the industry’s full development pipeline. Such high-profile early-stage projects are inherently highly uncertain and prone to delays.
Thus, the claim of “half the capacity delayed” applies only to niche speculative announcement projects and cannot represent the overall U.S. data center construction landscape or define the industry’s true outlook.
Delays are real—but attribution is clear and scope is controllable
SemiAnalysis does not deny the existence of delays in the sector and has preemptively warned about project risks for companies such as STACK Infrastructure, Oracle, Nebius, and Core Scientific. However, the institute makes it clear that industry delays are not systemic halts but concentrated in specific scenarios, dividing market delay cases into three core types.
- The first type is early-stage speculative projects, mostly launched by new industry entrants with aggressive planning and unrealistic delivery timelines, not yet at the stage of substantive implementation;
- The second type is timeline deviation in mature projects, where developers underestimate variables like equipment delivery, weather, and electrical commissioning and are overly optimistic about construction timelines;
- The third type is compliance-constrained delays, where projects in the later stages encounter approval obstacles, local public opposition, forcing design changes or new power sources, thus lengthening delivery cycles.
Taking Nebius’s New Jersey campus as an example, the first phase of 50MW was planned for 4 months but ultimately took 10 months to deliver, a typical second type delay. Core Scientific’s Denton campus, due to a mix of approval blocks, design changes, and supply chain disruptions, failed to meet the annual delivery goal of 250MW—a case of overlapping first and second types of delay.
Oracle/STACK’s New Mexico project is a classic third type of delay—gas pipeline obstruction and emissions approvals caused production start to be pushed from 2026 directly to 2029. All three cases point to the same conclusion: Delays are real, but their impact is relatively concentrated and the causal pathways are clear and traceable.
The essence of market panic: Three types of useless noise distort core judgment
From SemiAnalysis’s perspective, this round of systemic market panic stems from three types of “noise information” with no actual industrial impact, which have overly magnified the sector’s short-term risks.
The first is local construction suspension policies. As of April 2026, twelve U.S. states have issued regulatory bills, but only targeting early planning projects, with no effect on 2026’s core delivery capacity. The ban in Maine covers less than 5MW in planned capacity, without substantive impact.
The second is regional public sentiment resistance. Cases of protests, planning rejections, and company withdrawals mostly remain on paper, without land, equipment, or power grid agreements, so have never counted as deliverable capacity and do not affect real orders.
The third is a massive amount of invalid planning announcements. The power grid application queue shown includes a large volume of “virtual capacity.” For example, Texas ERCOT has received over 410GW of data center load applications (five times the historical peak), of which SemiAnalysis estimates about 311GW are speculative virtual demand that cannot be converted to real capacity.
Risk clearing, not demand collapse
SemiAnalysis believes that the fundamental cause of pessimistic misjudgment in the market is the failure to distinguish project risk layers.
This round of delayed or canceled projects is highly concentrated among early speculative circles—the structurally surplus part of the sector—whereas the core 2026 delivery capacity comes from high-quality projects with extremely high certainty, all possessing land rights, clear power supply solutions, solid approval progress, locked long-term equipment orders, and substantial construction progress.
For common industry bottlenecks, leading operators have developed mature solutions, with deep policies for local approval and community relations, pulling out early from high-resistance areas to policy-friendly zones or brownfield developments; hardware-wise, for grid access lasting 7–10 years and core equipment delivery cycles over a year, they use direct grid investment, pre-secured electrified land, distributed power, deposit-based equipment capacity locking, and modular prefabricated construction to effectively counter supply chain and infrastructure bottlenecks, ensuring core projects proceed on schedule.
Thus, panic stems more from risk misjudgment, while fundamental support remains unshaken.
Risk warnings and disclaimerThe market involves risks, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users' special investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Invest accordingly at your own risk.