US data center delays are crippling the AI build-out. Half of the 140 projects planned for 2026—representing 16 gigawatts of global capacity—face postponement to 2027 or outright cancellation, according to a Sightline Climate report published this week. The culprits are unglamorous but fatal: electrical grid equipment shortages, insufficient power capacity, and parts scarcity from China.
Key Takeaways
- 30-50% of 2026 data center projects delayed to 2027; only 5 GW under construction versus 11 GW in announced stage
- Cloud giants planned $650 billion in AI infrastructure spending for 2026, now threatened by supply and power constraints
- In 2025, 26% of 110 planned projects failed to launch on schedule, a pattern intensifying this year
- Local opposition blocked or delayed $64 billion in US data center projects over two years; 142 activist groups now operating across 24 states
- Transformers, switchgear, and batteries are the bottleneck—not compute chips or software
The Power Grid Cannot Keep Up
The AI boom demands electricity at a scale the US power infrastructure was never designed to handle. Olivia Wang, research analyst at Sightline Climate, framed the severity bluntly: delays are often the clearest public indication that a development is dead. Of the 16 GW planned globally for 2026, at least 11 GW remains in announced stage without any construction underway. Only 5 GW is actively being built, mostly in the US, leaving a massive gap between ambition and execution.
The problem is not theoretical. Developers need transformers, switchgear, and battery systems to connect data centers to the grid. These components are in short supply. China produces critical electrical equipment, and trade tensions have tightened availability. Utilities themselves lack the capacity to queue new projects—they are overwhelmed by the sheer volume of simultaneous requests. Developers often announce parallel projects to test permits and power availability, creating phantom load requests that confuse utilities and delay approvals even further.
QTS, one of the largest data center developers, announced a nearly 5 GW pipeline since January 2024. Yet about 25% of that capacity remains under construction despite 2025 online commitments. This is not a company failing to execute—it is an entire supply chain breaking under demand.
Why 2025 Projects Already Missed Deadlines
The 2025 cohort of projects was planned two to three years ago, before the AI acceleration became obvious. Even those comparatively easier conditions produced delays: 26% of 110 planned projects for 2025 were delayed or never built. Wang warned that 2026 projects face steeper headwinds. Labor shortages, equipment scarcity, and permitting backlogs have only worsened. If one-quarter of projects slipped under favorable conditions, expect worse from the 2026 slate.
The delays compound. Construction typically takes 12-18 months after permits are secured. If a project loses its 2026 window, it does not simply shift to early 2027—it cascades backward through the entire timeline. Utilities must upgrade substations and transmission lines before data centers can connect. Those upgrades take years. The grid is the constraint, not the data center builders themselves.
Local Revolt Blocks Billions in Investment
Beyond supply chains, communities are saying no. Data Center Watch tracked 142 activist groups across 24 states opposing new facilities. Over the past two years, local opposition blocked or delayed $64 billion in US data center projects—$18 billion canceled outright, $46 billion postponed. Residents fear higher utility bills, water depletion, noise, and property value decline. The opposition is not concentrated in one region; it spans from New York to Michigan to Virginia to Oklahoma.
GI Partners experienced a five-month delay on one project amid community pushback as of March 2025. Across 28 states, new local moratoriums on data center development have emerged in recent weeks, creating legal barriers that no amount of capital can overcome. The political resistance is bipartisan—neither conservative nor progressive communities want industrial facilities in their backyards, especially when the promised jobs are minimal and the power demands are enormous.
Cloud Giants Face a $650B Crunch
Cloud providers—Amazon, Google, Microsoft, Meta—planned to spend $650 billion on AI infrastructure in 2026. That figure assumed projects would launch on schedule. US data center delays mean some of that spending will shift to 2027 or disappear entirely. Competitors in other regions may capture share, but global delays are similar. The 16 GW planned for 2026 globally faces the same 30-50% delay rate as the US, driven by identical power and parts constraints.
This creates a strategic problem. AI model training demands vast compute capacity. If data centers are delayed, training timelines slip. If timelines slip, AI product launches slip. The companies betting on first-mover advantage in generative AI now face a supply chain bottleneck they cannot engineer their way out of. No amount of software optimization fixes a transformer shortage.
Is the US data center delay crisis temporary?
No. Olivia Wang stated that the pattern will persist and significantly heighten the likelihood of projects being delayed, withdrawn, or ultimately canceled. The power grid cannot be upgraded overnight. Electrical equipment manufacturing is a long-lead business—new factories take years to build. China’s dominance in components means US supply chains remain vulnerable to trade friction. Expect delays to worsen before improving.
How much of the $650B AI infrastructure spending will actually happen in 2026?
The research brief does not specify a revised spending forecast for 2026. However, if 30-50% of projects slip to 2027 and beyond, a proportional share of the $650 billion budget will shift with them. Conservative estimate: $200-325 billion of the planned 2026 spend will defer to later years.
Why are transformers and switchgear the real bottleneck?
Data centers consume power at industrial scale—a single facility can demand 100+ megawatts. The electrical infrastructure to deliver that power requires specialized, long-lead equipment. Transformers step down voltage; switchgear manages distribution; batteries provide backup. These components are manufactured globally, with China and Europe dominating production. Simultaneous demand from AI, renewable energy, and electrification has exhausted supply. Building new manufacturing capacity takes 3-5 years.
The AI infrastructure boom has collided with reality. Cloud giants have unlimited capital and engineering talent, but they cannot manufacture electrical transformers faster than physics and supply chains allow. US data center delays expose the unglamorous truth: the constraint is not innovation or investment appetite, but the mundane infrastructure that nobody noticed until it became the bottleneck. Expect this pattern to define AI infrastructure development for the next two years.
This article was written with AI assistance and editorially reviewed.
Source: Tom's Hardware
