AI's Data Center Boom Faces a Test of Power, Delivery and Debt
New reporting on delayed AI data centers puts attention on electricity access, financing and the gap between contracted demand and operating capacity.
The investment case for AI infrastructure is facing a more demanding test: whether developers can turn financing and customer commitments into facilities that operate on time and generate enough cash to support their costs.
A Financial Times report published Sunday, October 18, examined strains in the US data center expansion. It cited RegLayer research identifying at least 47 large projects delayed, withdrawn or blocked during 2026, and a Bernstein assessment that about 35% of the planned US pipeline was credible. Those are attributed assessments using different definitions, not interchangeable measures of completed capacity or an official industry census.
The new report adds to scrutiny of the physical and financial assumptions behind AI investment. It does not establish that demand for AI computing has disappeared.
A contract is one stage of the investment
Nscale's own financing disclosure illustrates the scale of commitments involved. On September 25, the company announced $3.36 billion in convertible financing and more than $103 billion in total contracted value. The financing comprised $2.36 billion at closing and an additional $1 billion Nvidia commitment expected to fund in mid-November.
Those figures measure different things. A future financing commitment is not cash already received. Total contracted value is not annual revenue, profit or the amount of customer cash immediately available to pay construction bills.
For investors, that creates several questions between an announcement and an operating asset: when funding arrives, when equipment is delivered, when the customer begins paying and what happens if either party misses a deadline. Large headline commitments cannot answer those questions without the associated terms.
Electricity projections show scale and uncertainty
The United States Data Center Energy Usage Report: 2025 Update, available through the Department of Energy, estimates that data centers could consume 11.8% of US electricity in its 2030 reference case. Its wider scenario range is 9.5% to 15.3%.
The report also estimates 2024 consumption at 192 terawatt-hours, or 4.7% of US electricity use. These are national estimates covering data centers generally, rather than AI facilities alone. They do not guarantee that any particular project will obtain a connection, and the forecasts depend on assumptions about equipment shipments, utilization and efficiency.
That distinction is central to the investment analysis. Nationwide demand growth can coexist with local construction delays or insufficient grid capacity. A facility needs usable power in a particular place at a particular time; an aggregate forecast cannot supply it.
TTN's explanation of AI data center electricity demand examines the relationship between computing expansion and power infrastructure.
Payment schedules can matter more than aggregate commitments
A project can have an attractive forecast over its full life and still face a shortage of cash before operations begin. Comparing the dates of equipment payments, interest charges and customer receipts helps identify that gap. A large multiyear customer commitment does not necessarily provide cash at the moment construction bills fall due.
The same distinction applies to capacity. Announced capacity, connected capacity and capacity accepted by a paying customer measure different stages. Treating them as interchangeable can overstate the near-term revenue base. Investors need compatible time periods and clear definitions before comparing spending with expected income, particularly when several companies participate in financing or delivering the same facility.
Delivery milestones deserve more attention
The most useful financial comparison is between money committed, capacity commissioned and cash earned. Each measures progress at a different stage. A developer can have substantial demand and still face a funding gap if spending precedes customer receipts for longer than planned.
TTN's assessment is that project selection and delivery deserve greater weight than a single sector-wide spending estimate. Evidence of energized capacity, customer acceptance and sustainable financing would strengthen an investment case. Repeated timetable revisions or dependence on unreceived funding would require closer examination. The current reporting raises those questions; it does not settle the economics of every AI infrastructure project.