Startupdaily
startupTelegram

AI's Debt Boom Is Starting to Look Like the Telecom Bubble

Venture Debt·November 20, 2026

AI's Debt Boom Is Starting to Look Like the Telecom Bubble

The race to build AI infrastructure is increasingly being financed with borrowed money, and one lender-side voice thinks the pattern should look familiar. In a guest column, David Spreng of Runway Growth Capital compares today's buildout with the telecom bubble of the late 1990s and early 2000s, when carriers and network builders took on heavy debt to lay fiber and add capacity well ahead of demand.

The core risk, in his telling, is timing. Data centers, chips, and power contracts are expensive, and the spending usually arrives long before the revenue it is meant to produce. Equity can absorb a slow ramp. Debt cannot. It comes with fixed payments on a fixed schedule, so if incoming cash trails outgoing payments, borrowers can be pushed into restructurings or bankruptcy, even if the underlying technology eventually works as promised.

Spreng's argument is not that AI will fail. He suggests the technology can deliver on its promise and borrowers can still get hurt. What separates survivors from casualties, he writes, is whether demand forecasts are realistic and whether debt is tied to dependable cash flows rather than optimistic adoption curves. A company that borrows against a best-case customer ramp is effectively betting on the calendar, and calendars are where lenders and borrowers both tend to get caught.

For founders and investors, the practical takeaway is to examine the financing structure as closely as the product. Useful questions include how much capacity is contracted before it is built, what happens if customer growth slips by a year, and whether repayment schedules assume revenue that has not yet arrived. The telecom era left behind plenty of useful fiber, but many of the companies that financed it did not survive long enough to benefit. That is the lesson the AI sector is being asked to remember.

Reporting based on an external source.