AI data center projects are delayed as power hookups fall behind

In short: Many AI data center projects are running late because it is taking years to get enough electricity connected, and more projects are also getting blocked by local opposition.

What's going on

Data center builders say the top reason projects fall behind schedule is access to power. Nearly half of industry professionals surveyed called power the biggest barrier. Some headlines frame this as “China holds the key,” but the bigger issue is that electricity grids, not one country, are struggling to keep up with demand.

In the United States, some data center projects face grid connection waits of up to seven years. That is like finishing a new restaurant but being told the kitchen gas line will not be ready until the next decade. In the UK, developers say they may need substation upgrades costing hundreds of millions of pounds.

Demand is also rising fast. OpenAI projects that have been publicly discussed could use 55.2 gigawatts of power, which is roughly enough for 44.2 million households. Deloitte has warned that the power used by AI data centers in the US could grow to 30 times today’s level within a decade.

What to watch

Delays are not only about electricity. In 2025, at least 25 AI data center projects were canceled, about four times more than the year before, and nearly 100 projects are being actively contested. Opposition includes environmental groups, rural landowners, and people worried about higher power and water bills, and nearly 40% of heavily contested projects fail.

China is taking a different approach by building power generation first, like large solar and wind sites in desert areas, and connecting data centers directly. That can reduce reliance on crowded national grids. Separately, builders say cooling is another bottleneck. 83% say local suppliers cannot meet demand for advanced cooling, and liquid cooling can cost 7% to 10% more than air cooling.

Source: Arstechnica