Key takeaways America's AI infrastructure ambitions approach $10 trillion At least 47 major projects face construction setbacks Unfinished data centers expose lenders and investors to losses
America's artificial intelligence boom is producing some of the largest infrastructure commitments in history. But an uncomfortable question is emerging behind the announcements: what happens to the billions of dollars promised to build data centers that never open?
An October 18 Financial Times investigation identified at least 47 major U.S. data center projects delayed, withdrawn or blocked during 2026. Bernstein analysts also estimated that only about 35% of the proposed development pipeline was credible, highlighting the difference between announced capacity and facilities likely to become operational.
The financial stakes are enormous. In a research paper, Columbia Business School professor Stijn Van Nieuwerburgh estimated that constructing 188 gigawatts of additional U.S. AI infrastructure by 2032 could require nearly $9 trillion.
That is projected investment, not money already spent. Nevertheless, the scale raises questions about who ultimately bears the cost when ambitious projects fail to materialize.
Why a $10 Billion Data Center Can Fail Before OpeningBuilding an AI data center involves far more than purchasing Nvidia GPUs. Developers must secure land, electricity, transformers, cooling equipment, regulatory permissions and enough financing to complete construction.
A company may announce a massive computing campus and sign preliminary agreements with prospective customers years before securing every component necessary to operate it.
The problem becomes particularly serious when promised electricity arrives years later than expected.
In September, Oracle issued a force majeure notice connected to Project Jupiter, a New Mexico campus developed through Blue Owl-backed STACK Infrastructure for OpenAI. The move exposed difficulties surrounding construction schedules and financing arrangements.
Similar problems have affected other proposed facilities, as local opposition, lengthy grid connections and rising construction costs complicate development.
Crucially, an announced $10 billion project does not mean $10 billion has already been borrowed or spent.
Who Loses Money When Construction Stops?The answer depends on how far a project has progressed and who financed each stage.
Before construction, developers may lose spending on land options, engineering, legal work and permits. If a project fails after financing closes, lenders and equity investors could face substantially larger exposures.
Large AI campuses are increasingly financed through special-purpose vehicles that own infrastructure and borrow against future lease payments.
Those arrangements can attract billions in financing without requiring the technology company purchasing computing services to own the physical data center.
But customer agreements are not necessarily unconditional guarantees. Contracts may include delivery deadlines, termination rights and other provisions that determine which party absorbs the financial consequences of delays.
That distinction matters for companies holding enormous AI contracts, because contracted revenue does not automatically translate into available cash.
