Six months ago, the biggest banks in the world were practically falling over themselves to finance the AI boom. Now they’re quietly trying to dump that same debt on anyone who’ll take it. And I mean anyone.

When the Music Stops, Find a Chair Fast

JPMorgan, Morgan Stanley, SMBC, and MUFG – names that don’t usually scream “nervous” – are shopping around pieces of their AI debt portfolio. We’re talking about the loans they made to fund data centers for Oracle, Meta, and the rest of the companies building out what’s supposed to be the future of computing. The banks aren’t just trying to sell this stuff. They’re exploring what’s called “significant risk transfers,” which is finance-speak for “we’ll keep this on our books technically, but please, for the love of god, take the actual risk off our hands.”

Here’s the thing – these banks have hit their internal limits. They’ve lent so much money to AI-related projects that their own risk departments are basically screaming stop. And when JPMorgan gets skittish about tech lending, you should probably pay attention.

But the real story isn’t just that they’re trying to sell. It’s how hard they’re trying, and how stuck they’ve gotten.

$38 Billion Nobody Wants to Touch

There’s this one deal that really tells you everything you need to know. A group of banks including JPMorgan and MUFG have been sitting on $38 billion in construction debt – money they lent to build data centers in Texas and Wisconsin that Oracle’s supposed to lease. They’ve been trying to offload this thing for more than six months now.

Six. Months.

These aren’t loans you underwrite planning to hold forever. Banks make money by moving debt around, packaging it, selling it to investors who want the yield. That’s the whole model. But apparently nobody wants this particular pile of yield badly enough to take it at face value. So now some of these lenders are willing to sell pieces at a discount just to get it off their balance sheets.

Think about what that means. Banks are essentially saying “we’ll take a loss right now rather than keep holding this.” That’s not normal market friction. That’s concern.

When a Hedge Fund Manager Says “Out of Scale”

Matthew Moniot runs credit risk sharing at Man Group, which means his entire job is evaluating and buying exactly this kind of bank debt. And here’s what he told the Financial Times about the sizes of these AI deals: they’re “out of scale to anything we’ve thought about, ever.”

“Out of scale to anything we’ve thought about, ever.”

That’s not some random analyst being dramatic. This is a guy whose firm manages over $170 billion, who looks at massive debt transfers for a living, saying he’s never seen anything like this. And look, I’ve covered enough of these cycles to know that when the sophisticated money starts using phrases like that, it’s not because they’re impressed. It’s because they’re worried about what happens if this goes sideways.

The Timing Is… Interesting

What makes this especially wild is the speed of the reversal. Six months ago, these same banks were competing to underwrite AI infrastructure loans. The pitch was simple: AI is the next internet, data centers are the new real estate, get in now or miss the boom. Every bank wanted a piece because the fees were huge and the narrative was bulletproof.

But narratives change fast when risk committees start asking uncomfortable questions. Questions like: what if these data centers cost more to build than projected? What if AI demand doesn’t grow as fast as everyone’s betting? What if – and this is the big one – what if Oracle or whoever else is supposed to lease these facilities decides the economics don’t work anymore?

What This Actually Means

I’m not gonna sit here and tell you the AI boom is over or that data centers are a bad investment. I don’t know that, and neither does anyone else right now. But I do know that when the banks closest to the money start hedging their own bet this aggressively, that’s information. Real information. It's worth noting that over 400 scientists want AI unplugged – now, which might tell you something about what the smart money actually thinks when the people building the technology start pumping the brakes.

The banks aren’t bailing because they think AI is fake or because data centers won’t matter. They’re bailing because they’ve lent too much, too fast, to too few counterparties, and their risk models are flashing red. They got caught up in the momentum just like everyone else, and now they’re trying to get out before they’re overexposed to a sector that might not pay off as quickly – or as reliably – as everyone assumed six months ago.

And if the lenders who arranged these deals can’t find buyers at par? Well. That tells you something about what the smart money actually thinks is going to happen next. Not what they say in conferences or pitch decks. What they’re willing to pay, right now, in private.

The AI buildout isn’t stopping. But the easy money part? That might already be over.