Two analysts debate whether the AI boom could trigger a crash

In short: The AI boom is real, but some analysts warn that the way it is being paid for could still end badly.

What's going on

Azeem Azhar and Paul Kedrosky both say AI is a major technology shift. Where they differ is on the financial risk around today’s big buildout of AI infrastructure.

Kedrosky argues that companies are borrowing heavily to build data centers, meaning large buildings packed with computers that run AI (like building new power plants before you know how much electricity people will buy). He compares parts of this to the run-up to the 2008 financial crisis. His concern is simple, if the money made from AI does not grow fast enough, borrowers may struggle to pay back loans, and lenders and investors could take losses.

He also worries about what he calls a “slop economy.” In plain terms, he thinks AI could produce a huge amount of software, text, images, and other output that looks useful, but does not create enough real economic value to justify the spending.

Azhar counters that there is real demand and real growth. He estimates AI sector revenue at about $110 billion in the 12 months to June 2026, which he puts at roughly 0.4 percent of US GDP (the total size of the US economy). That suggests fast growth, but also that AI is still small compared to everything else. He also says signs of credit stress are a warning, not proof that a crash must happen.

What to watch

The big test is whether customer revenue and clear productivity gains catch up with the enormous cost of data centers. Kedrosky points to refinancing risk later in the decade, meaning some loans will need to be replaced with new ones, possibly at higher rates. Azhar describes the situation as finely balanced, with outcomes still uncertain.

Source: NYTimes