AI’s infrastructure boom is growing more dependent on debt

AI infrastructure spending is accelerating, and the companies building data centres are increasingly borrowing to keep up, argues Noah Smith in a 15 October essay. His point is not that a crash is imminent, but that the boom’s financial exposure is greater if AI revenue growth slows.

Watch Desk analysis

What happened

Smith contrasts today’s AI build-out with the US railroad boom of the 19th century. He argues that early warnings about an AI infrastructure bust looked less pressing when hyperscalers were mostly paying for data centres from cash flow. That picture has changed as spending has grown and borrowing has become more important.

The essay says AI revenue has also accelerated, in part because of demand for coding tools such as Claude Code and Codex. But Smith warns that debt raises the amount of revenue companies need to sustain their investment. He notes that estimates of the AI boom’s share of GDP vary substantially, partly because analysts count different costs and kinds of infrastructure. The railroad comparison is interesting, but the figures are not settled enough to make it a tidy historical match.

Why it matters

The argument connects the software people are buying with the infrastructure being financed to run it. If demand for AI products keeps growing, borrowing may help builders expand capacity. If growth falls short, debt could make the consequences sharper for companies that have committed heavily to data centres and related infrastructure.

That is a risk scenario, not evidence that a bust is underway. Smith’s essay presents a case for why the stakes have risen, rather than a forecast with a firm timetable or a single agreed measure of the boom.

Our read

The useful question is no longer simply whether AI will prove valuable. It is whether the revenue arrives quickly and reliably enough to justify the infrastructure being built for it. Smith’s railroad analogy is a vivid comparison, but the more immediate signal is the shift from spending cash to taking on debt. The boom may be durable; its financing is becoming harder to shrug off.

What to watch

• Whether AI-related revenue continues to grow alongside infrastructure spending.
• How much hyperscalers borrow, and how their cash flow develops.
• Whether analysts converge on more comparable estimates of the scale of the build-out.

Discussion spark: If AI revenue growth slowed, would the greater risk lie with the companies borrowing to build data centres, or with the wider technology industry depending on that spending?

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.