Blackstone Retreat: Is the US AI Data Center Construction Overvalued by the Market?

Recently, Blackstone-backed data center operator QTS announced the official abandonment of the Digital Gateway project, which had planned to build the world's largest data center campus with an investment exceeding $100 billion. After years of resident opposition, legal disputes, and approval setbacks, coupled with the earlier withdrawal of partner developer Compass Datacenters in May, QTS ultimately decided to halt the project.

This is not an isolated case. Constrained by multiple factors such as power supply, approvals, land acquisition, community resistance, and supply chain bottlenecks, a large number of AI data center projects continue to face delays, with actual construction progress significantly lagging behind market expectations. In Q1 of this year, at least 75 data center projects in the U.S. (worth about $130 billion) were canceled or delayed, approaching the total for the entire previous year. Meanwhile, over 60% of data centers planned for commissioning before 2027 have not yet officially broken ground, and another 7% face explicit delays.

At the same time, tech giants like Microsoft, Google, Meta, and Amazon continue to raise their capital expenditure budgets (projected at $725 billion in 2026), driving ever-higher market expectations for AI computing power demand and data center construction scale. However, this linear extrapolation based on model-side demand significantly overestimates the actual delivery capacity of data centers in the physical world. Once the construction phase encounters obstacles, the hundreds of billions of dollars in CapEx from tech giants will be difficult to convert into effective computing power assets as planned, making a supply chain valuation re-evaluation inevitable.

Furthermore, as the AI data center construction cycle lengthens, the cash flow recovery period for many projects relying on debt financing continues to shift backward, leading to a rapid accumulation of asset depreciation, financing costs, and leasing risks. AI infrastructure will gradually shift from an industry logic to a financial logic.

CapEx does not equal new computing power. When will the market re-evaluate AI investment efficiency?

Over the past two years, a relatively simple investment logic in the capital market has been: AI demand continues to grow, tech giants keep raising CapEx, data center construction accelerates accordingly, and companies in the supply chain benefit continuously. Therefore, the scale of CapEx itself has become the most important observation indicator for the market.

But CapEx does not equal new computing power. Data centers are typical heavy-asset infrastructure, typically requiring several years from project initiation, approval, land acquisition, grid connection to official operation.