JPMorgan Just Quietly Marked Down Its AI-Linked Loans

While the world was watching oil prices spike and bombs fall in the Middle East, JPMorgan Chase did something that barely made a ripple in the headlines — but should have. The nation’s largest bank quietly began marking down the value of loans tied to private-credit portfolios, many of them extended to software companies. When the biggest bank in America starts adjusting collateral values in one of the hottest lending markets on the planet, it’s not a clerical footnote. It’s a signal.

Here’s why this matters. Private credit has exploded into a multi-trillion-dollar industry over the past decade. After regulators clamped down on bank lending following 2008, private lenders rushed in to fill the gap — offering flexible loans to companies traditional banks wouldn’t touch, in exchange for juicy yields. The system works beautifully as long as borrowers keep generating cash. But a growing slice of those borrowers are now pouring money into AI infrastructure — data centers, cloud buildouts, compute capacity — where the upfront costs are staggering and the profits are still mostly theoretical.

Take Oracle as a case study. The company recently signed a $300 billion cloud deal to deliver 4.5 gigawatts of computing power to OpenAI between 2027 and 2032. Building that out costs roughly $225 billion — $35 billion per gigawatt in Nvidia chips alone, plus another $15 billion per gigawatt in supporting infrastructure. On paper, the $75 billion spread looks attractive. In practice, the margin for error is razor-thin. If a single major customer delays, if pricing pressure emerges from competitors offering functionally identical Nvidia-powered services, or if AI demand simply doesn’t materialize at the pace everyone is betting on, those economics unravel fast.

And the stress signals aren’t just coming from JPMorgan. Private credit giant Blue Owl Capital recently faced a surge of redemption requests in one of its funds, forcing the firm to restrict withdrawals and liquidate roughly $1.4 billion in loans to raise cash. While bad AI loans weren’t the direct trigger, it highlights a fragile truth: a massive share of private-credit lending today is going to software companies whose own business models could be disrupted by the very AI revolution they’re financing.

This is the part of the AI story that doesn’t get enough attention. Everyone is focused on which chipmaker or cloud provider will “win” the AI race. But the real risk may be hiding in the debt markets propping the whole thing up. When JPMorgan — the smartest risk managers on Wall Street — starts quietly reducing exposure, investors should at least pause and ask what they know that the rest of us don’t. The AI infrastructure buildout isn’t slowing down, but the assumption that every dollar spent will generate a profitable return is starting to crack. And in markets, cracks have a habit of widening before anyone expects them to.