America’s AI data-centre boom may require nearly $9 trillion in investment by 2032, while dozens of proposed projects are already delayed, withdrawn or blocked. The question is not just how much gets built, but who pays when a campus never opens.
Watch Desk analysis
What happened
Coinpaper reports that an October 18 Financial Times investigation identified at least 47 major US data-centre projects facing setbacks in 2026. Bernstein analysts estimated that only about 35% of the proposed development pipeline was credible. Those are reported estimates, not a tally of completed projects or money already lost.
A research paper by Columbia Business School professor Stijn Van Nieuwerburgh puts the investment needed to construct 188 gigawatts of additional US AI infrastructure by 2032 at nearly $9 trillion. Developers must line up land, electricity, transformers, cooling equipment, permits and finance before a facility can operate. An announced project’s price tag is not the same as money borrowed or spent.
Why it matters
The risks change as a project advances. Before construction, developers may lose money on land options, engineering and permits. Once financing closes, lenders and equity investors can face much larger exposure. Special-purpose companies may own and borrow against data-centre assets, while the AI firm buying computing capacity does not own the site itself.
Customer agreements do not automatically guarantee payment in every circumstance. Delivery deadlines, termination rights and other contract terms can determine who absorbs losses when a build is late or abandoned. Coinpaper points to Oracle’s September force majeure notice connected to Project Jupiter, a New Mexico campus developed by STACK Infrastructure for OpenAI, as an example of construction and financing difficulties coming into view.
Our read
The useful distinction is between an enormous AI infrastructure commitment and a functioning data centre with customers, power and a clear route to payment. The first can make a striking headline; the second is what lenders need. Before treating projected investment as a settled boom, readers should ask how much of the pipeline is actually financeable and what the contracts say if delivery slips.
What to watch
- Whether delayed or blocked projects secure the power, permits and finance needed to proceed.
- Which parties bear costs when delivery deadlines are missed or contracts are terminated.
- Whether the estimated 35% of credible projects becomes actual operating capacity.
Discussion spark: When an AI data-centre project fails to open, should the main financial risk sit with its developers and lenders, or with the technology customers whose demand helped justify the build?
Sources and evidence
- Source update (18 October 2026, 14:02 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.