The rapid expansion of AI data centers in the United States is stalling, with nearly half of planned projects now facing delays or cancellation.
In recent years, artificial intelligence has become one of the most widely discussed and heavily invested-in sectors globally. Its impact is already reshaping industries ranging from the sciences to the creative arts, and experts continue to warn that the next two years could bring technological and societal changes on an unprecedented scale.
It is no wonder that this momentum has caused billions of dollars to be poured into new AI models and infrastructure. Advances in AI have created a surge in the construction of data centers —high-performance facilities packed with interconnected processors, specialized chips, and vast data systems that require an immense amount of energy to power AI.
Leading tech companies such as Amazon and Microsoft are projected to spend more than $650 billion in 2026 alone on AI development and the buildout of these so-called “AI factories.”
Despite the hype and vast amounts of capital, however, a recent Bloomberg report indicates that nearly half of all planned U.S. data center projects have been delayed or canceled. Additionally, more than 60% of projects slated for completion in 2027 had not begun construction as of April 2026.
The primary hinderance appears to be supply chain constraints. Critical components—including transformers, circuit breakers, and industrial-scale batteries—are in short supply, yet these are essential to build and operate these large data centers.
Despite years of U.S. reshoring efforts, American manufacturing capacity for these components is simply not enough to keep up with the demand.
As a result, companies still rely heavily on imports. Canada, Mexico, and South Korea are key suppliers of high-capacity transformers, while imports from China have reportedly surged more than 400% over the past three years.
At the same time, the surge in demand for these specialized parts has outpaced the ability of manufacturers to produce and assemble them quickly enough. The time required to manufacture these components is notoriously long. Shipping delays, tariffs, and geopolitical tensions—particularly involving China—have further complicated logistics.
Zhao Haijun, co-chief executive of Semiconductor Manufacturing International Co. in China, has warned that the U.S. may face a similar outcome to China’s recent experience. Following China’s 2023–2024 data center boom, nearly 80% of newly built facilities remain underutilized or vacant, highlighting the risks of overexpansion and under-planning.
