The safest market in the world is priced for zero risk. One name isn't.

I like boring.

In my own account, the boring part is the corner I never check. The investment-grade bonds. The adult money that just sits there and pays me.

You give up excitement for calm. That's the whole deal.

But the calm is what worries me now.

The investment-grade market — thousands of the safest corporate bonds in America — is priced for almost no risk at all. The extra yield those bonds pay over Treasuries sits near 77 basis points.

That's about the lowest since 2007. The market last felt this safe right before the last crisis.

Now look at one name inside all that calm. Oracle.

A 48-year-old software company. About as dull as they come, once.

The cost to insure Oracle's debt against default has climbed to about 198 basis points — a record. More than double where it sat last autumn.

So the market says two things at the same time. In aggregate: nothing to see here. On one of its largest borrowers: get me protection, now.

That gap is the story.

And Oracle isn't alone. It's the loudest voice in a much bigger room.

When one company's default insurance doubles while the whole market yawns, I want to know why. When tech giants move a hundred billion dollars of debt off their own balance sheets, I want to know where it went.

So this week I followed the money behind the AI buildout — the borrowing that never shows up on a balance sheet — and asked one simple question.

If the safest market in the world is this calm, who is carrying the risk?

The AI buildout costs more money than even the richest companies on earth keep on hand. So they borrowed. But they did it in a way that keeps most of the debt out of sight.

The move is simple. A tech company sets up a separate legal entity — a special-purpose vehicle. The vehicle owns the data center: the land, the buildings, the chips. Wall Street and private funds lend the vehicle the money. The tech company just signs a long lease.

The debt sits in the vehicle. Not on the parent's balance sheet.

The Financial Times counted more than $120 billion of AI data-center spending moved off the books this way, by Meta, Oracle, xAI and CoreWeave.

Meta's Louisiana project shows how it works. A vehicle set up with Blue Owl raised about $27 billion in loans and $3 billion in equity. None of it landed on Meta's balance sheet. Weeks later, Meta raised another $30 billion in ordinary bonds, with the first pile of debt nowhere in sight.

The rating stays clean. The risk doesn't go away. It just moves somewhere harder to see.

And here's where it spreads. Somebody funded those vehicles. More and more, it's private credit — investment funds that lend outside the banking system, with far less disclosure.

Loans from private funds to AI companies have gone from near zero to more than $200 billion in a few years. Morgan Stanley thinks private credit will supply another $800 billion for data centers over the next two years.

That money doesn't come from nowhere. It comes from the pensions, insurers and banks that back the private funds.

So the chain runs like this. Your pension fund lends to a private-credit fund. The fund lends to a data-center vehicle. The vehicle collects a lease from an AI company that may or may not turn a profit.

Private credit's whole appeal is that it doesn't trade every day, so the marks stay smooth. The catch is the same thing. If a loan sours, you may not find out until the loss has already traveled up the chain.

Meanwhile, the banks lending to these giants have run into an old limit. A bank can only carry so much exposure to one borrower. And a handful of AI names now soak up enormous amounts of credit.

So the banks are hedging. They buy protection on the very companies they lend to.

Trading in default swaps on Microsoft, Amazon and Oracle hit $4.6 billion in the first quarter of this year. A year earlier it was $759 million. Six times more, in twelve months.

On the other side of some of those trades sits Boaz Weinstein's Saba Capital, happy to sell the protection the banks want.

One number frames the strain. Oracle's debt now runs to more than four times its equity. Its hyperscaler peers sit under half of theirs.

When lenders start insuring themselves against their best customers, it's worth asking what they see.

Three more things worth keeping track of…

Japan ·  Japan is struggling to sell its long bonds. The yield on the 40-year Japanese government bond touched a record above 4% this year, a first for any Japanese bond in more than three decades. The government's answer was telling. This month the finance minister asked the country's own pension funds to buy more Japanese assets. When a government has to lean on its own pensions to hold yields down, the open market for its debt is getting thin.

Gold ·  Gold crossed a line most people missed. For the first time since 1996, gold makes up a bigger share of the world's central-bank reserves than U.S. Treasuries, according to Morgan Stanley. And central banks keep buying more even with the price above $5,000, the opposite of how a normal buyer behaves. They aren't trading. They're stepping away from paper money.

Cars ·  The bottom of the consumer is straining. More than 6.8% of subprime car loans ran at least 60 days past due at the start of 2026, the worst in 32 years, per Fitch. Lenders repossessed 1.73 million vehicles last year, the most since 2009. Prime borrowers are fine. The pain sits entirely with the people who can least handle it. We'll see.

You've met this person. The friend who lends a buddy money so the buddy can pay him back what he already owes. It feels like generosity. It's really the same dollar going in a circle and getting called income.

Companies do it too. Right now the biggest one is Nvidia. It agreed to invest up to $100 billion in OpenAI. OpenAI then spends much of that money buying Nvidia's chips.

AMD ran its own version. It handed OpenAI the right to buy up to 160 million AMD shares in exchange for a promise to run AMD chips.

"Most of the money will go back to Nvidia." — OpenAI's finance chief

We've seen this movie. In 1999, the hottest companies on earth sold telecom gear. Lucent. Nortel. Cisco. Their customers were new phone and internet carriers with big dreams and empty pockets.

So the equipment makers became the bank. Lucent committed about $8 billion in loans to its own customers. Nortel added a few billion more. Cisco chipped in too.

The customers used the loans to buy equipment. The makers booked the sales. Revenue soared. Everyone looked brilliant.

Then the customers ran out of road. Forty-seven of those carriers went bankrupt between 2000 and 2003.

The bill came back to the lenders. Lucent lost $16 billion in 2001 alone, more than every dollar of profit it had ever made. Its stock fell from $65 to 76 cents. Nortel went from a $136 billion company to a $14 billion one in two years.

Near the end, they weren't selling equipment. They were giving it away and calling it a sale.

The difference this time, everyone says, is that the customers can pay. Microsoft and Amazon and Google earn enormous profits. Maybe that holds.

But the shape is the same. A seller lends you the money to buy its product, and the sale looks like proof the demand was there all along.

It's a great trade while the circle keeps spinning. The trouble starts the day someone needs the dollar to stop.

We'll see.