Nearly half of all U.S. data centers planned for 2026 are facing delays or outright cancellation. The culprits? Power constraints, supply chain bottlenecks, and local opposition. Everyone's building AI models, but nobody can build the datacenters to run them fast enough.
The numbers are stark. Industry analysts estimate that 47% of planned facilities won't meet their original construction timelines. Some projects are pushed back 12-18 months. Others are quietly being shelved as companies realize the infrastructure challenges are insurmountable at current scale.
Power is the biggest constraint. A modern AI datacenter needs dedicated substations, reinforced transmission lines, and often entirely new power generation capacity. Getting utility companies to commit takes years, not months. By the time permits clear, the AI models you planned to train might already be obsolete.
Supply chains aren't helping. High-performance GPUs from Nvidia and AMD are backordered for quarters. Specialized cooling systems—required because AI chips run ridiculously hot—have 8-12 month lead times. Even basic construction materials like concrete and steel are allocated to competing megaprojects.
Local opposition is the wildcard. Communities that initially welcomed datacenters for tax revenue are now pushing back on water usage (for cooling), noise pollution (from generators and HVAC), and electrical load (which can spike local utility rates). Virginia, Arizona, and Texas—all datacenter hotspots—have seen organized resistance to new facilities.
This infrastructure bottleneck might be the real limiting factor for AI deployment, not the algorithms themselves. OpenAI can design GPT-6, but if they can't get the compute to train it, the model stays theoretical.
Companies are getting creative. Some are buying existing industrial facilities and retrofitting them. Others are exploring international locations where regulations and power access are more favorable. A few are even talking about building their own nuclear reactors, which tells you how desperate the situation is.
The AI boom is real. The infrastructure to support it isn't catching up.
