“We’re doing it again,” Venture capitalist Paul Kedrosky said on the Between Two COOs podcast. Exorbitant debt to finance the AI infrastructure boom reminds him of the run-up to the 2008 financial crisis, but on a “vastly larger scale.”
The result, he warned, is a “slop economy” where companies can produce vastly more software, presentations and other content, even when little additional value is created.
“We’re producing vast amounts of everything, but for the most part, nobody cares,” Kedrosky
More Apps, More Decks, but More Productivity?
Kedrosky pointed to an NBER study covering more than 500,000 GitHub developers. Autonomous coding agents increased coding activity by 240%. That gain fell to 80% at the project level and 30% for actual releases.
In other words, new software applications aren’t translating into increased usage.
Nowadays, junior bankers use AI to produce “hundreds more investment decks” and pitch transactions that may never happen. “It’s not clear there’s a strong rationale for why it’s being done in the first place,” he added.
AI’s ‘Token Factories’ Meet a Debt Boom
Big-spender Amazon.com Inc. (NASDAQ:AMZN) expects capital expenditures of about $220 billion this year as it races to add AI and cloud capacity. That buildout fuels demand for Nvidia Corp. (NASDAQ:NVDA), although Amazon is also investing heavily in its own AI chips.
Prediction traders expect the spending boom to continue. Polymarket currently gives Amazon a 61% chance of 2026 capex exceeding $220 billion.
Kedrosky’s concern isn’t whether demand is strong today. It is whether the data centers being built can eventually earn enough to justify the money pouring into them.
Kedrosky estimates more than 60% of AI financing is now debt-backed, up from roughly 15% to 20% a year ago.
Earning those returns becomes harder if the product those data centers sell keeps getting cheaper. Kedrosky describes them as “token factories” producing what he calls a “hyper-deflationary industrial commodity.”
‘There’s No Economics That Does That’
In a separate Better Offline interview, Kedrosky said higher Treasury yields and credit spreads are pushing required returns on some stressed data-center projects toward 10% to 12%.
“We’re within six to 12 months on the outside of this … all breaking,” he said. “There’s no economics that does that.”
Polymarket traders currently put the probability of an AI industry downturn by Dec. 31 at about 10%, suggesting traders see relatively little risk of the contract’s conditions being triggered before year-end.
Kedrosky sees another test approaching later. He expects some of the five-year financing structures behind the AI buildout to hit a refinancing wall around 2029 and 2030.
"Will it all get refinanced? No freaking way," he said. "And then if it does get refinanced, it’ll be refinanced at prohibitive terms."
But don’t call Kedrosky an AI skeptic. “It’s a tremendously important technology,” he said on Between Two COOs. “Probably the most consequential of my lifetime.”
Image: Shutterstock
Login to comment