AI's Biggest Risk Isn't Technology—It's Credit

Off-balance-sheet debt, private credit, and AI infrastructure are converging into one of today's biggest market risks.

KEY TAKEAWAYS

• The AI infrastructure boom is increasingly being financed through complex credit structures rather than traditional corporate balance sheets. This shifts risk away from visible debt and into private markets and special-purpose entities.

• The Bank for International Settlements has elevated AI-related financial vulnerabilities alongside inflation and sovereign debt as three major macroeconomic pressure points. That framing suggests AI financing has become a systemic issue rather than simply a technology story.

• Off-balance-sheet lease commitments tied to hyperscaler infrastructure have grown to enormous levels. While these obligations comply with current accounting standards, they still represent meaningful future economic commitments.

• Private credit has become a critical funding source for AI infrastructure. A slowdown in AI investment or monetization could ripple through contractors, lenders, refinancing markets, and ultimately the broader financial system.

• Investors should focus on leading indicators such as hyperscaler capital spending, free cash flow, credit spreads, and private credit fundraising rather than reacting after market stress appears.

MY HOT TAKES

• Technological revolutions rarely create the greatest financial damage by themselves; excessive leverage built around those revolutions often does.

• Visible debt is usually manageable because markets can price it. Risks hidden within financing structures deserve far greater attention.

• The AI investment cycle remains fundamentally constructive, but investors should distinguish between enthusiasm for the technology and complacency about its financing.

• Central bank reports often provide important early warnings long before markets fully appreciate emerging systemic risks.

• Successful investing requires maintaining a wide-angle perspective, identifying developments occurring beneath the market's most popular narratives.

• You can quote me: "Credit problems don't begin with headlines. They begin in the footnotes."

Eyes wide. I know it’s not trendy anymore–it may even be socially taboo–but when I was young an annual visit to the circus was an institution. Where I grew up, the circus would come to town early spring, every year. It was a spectacle with three rings. Sometimes during the show all three rings would be active, challenging the audience to focus on the main spectacle. If you were too focused on the clown car (dozens of clowns somehow would fit into a tiny car), you might miss the lion tamer’s wife sticking her head into the jaws of a ferocious lion, or an elephant dancing on its hind legs with a french poodle. If you looked down at your Cracker Jacks box, you might miss the never-attempted, death defying triple-flip-twist on the trapeze that hung way up high in Madison Square Garden (we call it simply, “the Garden”).

When I wake up each day and look at the markets and the economy, I try to start from the view of the “cheap” seats. The ones that are so high up in the Garden that you have no choice but to have a wide view–you see everything. I have to say that these past 18 months or so have been quite the three ring circus with so much to see that it might be easy to miss the lady spinning on rope held in her teeth. We have geopolitical hotspots all over the globe, an increasingly hawkish Fed with a new ringmaster, a polarizing mid-term election on the horizon, a stock market trading near all-time highs, artificial intelligence promising to disrupt (positively and negatively) everything, the bond markets running amok–that’s just the start of it.

The hard part is recognizing activity that might provide some hints on what’s next. To recognize that the cannon that will soon launch a man across the Garden is being rolled out in the ring with the dimmed spotlights, you would have to ignore the knife thrower pierce an apple sitting on his beautiful assistant’s head–amidst a flaming ring of fire.

My friends, the other day I noticed the cannon being set up in ring number two under dimmed light. To be fair, I was expecting it at some point. In fact, I began bringing it to your attention last year when I stated my concern about a debt problem that would be hard to recognize because it was not debt as we know it. You wouldn’t hear Jamie Dimon or any of his titan-banker colleagues talking about it. No, this was the stuff that happened between private equity lenders, large institutions, sovereign funds, and lawyers–lots of lawyers. That last one should give it away. 😉

That debt problem I flagged last year? It literally just got its own chapter in the institution’s report that every other central bank treats as gospel. On June 28th, the Bank for International Settlements–the central bank for the world’s central banks, the place where the Fed, the ECB, and the Bank of England all compare notes–released its 2026 Annual Economic Report. And buried in there, dressed up in careful central-banker language, is a warning that the AI infrastructure boom has built itself a debt structure most investors can’t see, don’t understand, and absolutely should not ignore.

Now, before you assume I’m here to tell you the sky is falling–I’m not. I never am. That’s not how I do things, and frankly it’s not useful. The lion didn’t actually eat the trainer’s wife. The trapeze artist always sticks the landing–that’s why we keep buying tickets. But the smart audience member, the one in the cheap seats with the wide view, watches the cannon get rolled into position anyway. Awareness isn’t alarm. It’s just paying attention to the parts of the show that aren’t lit up.

Here’s the official narrative, and to be fair, it’s not wrong, it’s just incomplete. The five biggest hyperscalers–Alphabet, Amazon, Meta, Microsoft, and Oracle–are on pace to spend over a trillion dollars combined on AI infrastructure capital expenditures across 2025 and 2026. The BIS itself acknowledges that this spending helped prop up global growth last year even as tariffs and geopolitical noise threatened to drag things down. So far, so good. This is the version of the story that gets you a segment on the morning shows and a green arrow next to your favorite tech ETF.

But the BIS report doesn’t stop there, and this is where it gets interesting for those of us who actually read past the executive summary. 🙋 The report frames the current moment as one of three connected “pressure points” facing the global economy right now. AI-related financial vulnerabilities sitting right alongside stubborn inflation and ballooning sovereign debt. That’s not a tech analyst’s opinion. That’s the institution that literally exists to keep the plumbing of global finance from bursting, telling you these three things are now wired together.

Let’s talk about the plumbing itself, because that’s where the real story lives, not in the splashy capex headlines, but in how all that capital actually gets raised. Hyperscaler bond issuance crossed the $100 billion mark in 2025 alone, more than four times the five-year average for these companies. That’s debt you can see, sitting right there on the balance sheet for anyone with a Bloomberg terminal and ten minutes to spare. I don’t lose sleep over visible debt. Visible debt gets priced, rated, and traded by people whose entire job is pricing, rating, and trading debt.

What I lose sleep over–what should make you set down your cotton candy for a second–is the debt that doesn’t show up there at all.

Here’s how the structure works, and once you see it you can’t unsee it. A hyperscaler doesn’t always build its own data center and put the debt on its own books. Instead, it spins up a special purpose vehicle (SPV) or a joint venture. That entity goes out and either acquires or develops the data center asset. Private credit shops–and we’re not talking about anybody’s neighborhood bank here, we’re talking Blackstone, Apollo, Blue Owl, PIMCO, BlackRock–step in and fund the debt through private placements, often investment-grade rated, often anchored by billions of dollars from a single lender. The hyperscaler then signs a long-term lease to use the facility. Voila–what used to be a capital expenditure on the company’s own balance sheet quietly becomes an operating lease instead. The debt sits in the SPV. The hyperscaler’s balance sheet stays pristine. Everybody’s metrics look great.

Moody’s actually did the math on how big this gap has become, and the number stopped me cold. Their accounting team calculated that across just five hyperscalers, there’s $662 billion in lease-related commitments that aren’t recorded as liabilities on any of their balance sheets right now. To put that in perspective, that’s roughly 113% of those same five companies’ most recent adjusted debt, meaning the stuff hiding off the books is actually bigger than the stuff sitting on it. Meta alone has a $28 billion residual value guarantee tied to one of its data center arrangements–a promise to make investors whole if the facility’s value comes in under a certain threshold years from now–and because management deemed that outcome “not probable,” it shows up nowhere except a footnote. Not a lie. Not fraud. Just accounting rules written for a world that didn’t anticipate AI infrastructure moving this fast, this big, this opaque.

And look, I want to be fair here, because Moody’s own analysts made a point I think gets lost in the scarier headlines–this isn’t “hidden” debt in the sinister sense. Nobody’s cooking the books. It’s debt that hasn’t technically triggered yet under current accounting standards, but that for all practical economic purposes, exists. These companies haven’t avoided a liability through clever structuring so much as they simply haven’t yet received the service that would formally trigger it. That’s a more honest framing than “secret debt.” But “not yet triggered” and “not a risk” are two very different sentences, and the BIS is the one institution with the standing to say so out loud.

Let’s sit with that Moody’s number for one more second. 113% of adjusted debt, sitting off balance sheet, across just five companies. If you’d told me twenty years ago that the most important number for understanding tech sector risk wouldn’t be in the debt-to-equity ratio at all, I’d have laughed you out of the room. Today it might be the single most important number nobody on financial television is talking about.

The BIS flagged something else that deserves your full attention, what it calls circular financing. This is the part of the circus where the same act seems to be happening in two rings at once, and if you’re not watching closely you’d swear it was two different performances. Chipmakers and hyperscalers take equity stakes in AI labs. Those AI labs, in turn, commit to multi-year purchases of chips and compute, often from the very same companies that just invested in them. On paper, the revenue is real. The contracts are binding. Wall Street analysts plug those numbers into their models and the growth story writes itself. But when you actually trace where the dollars originate and where they end up, a meaningful share of it is just recycling within the same small ecosystem of players. That’s not organic demand pulling the AI buildout forward. That’s manufactured demand, and manufactured demand has a way of flattering every metric right up until the moment it doesn’t.

Now here’s the transmission mechanism: the part that should matter most to anyone with money in the market, not just the people watching credit spreads for a living. If AI monetization disappoints. If the revenue these massive capex budgets are betting on doesn’t show up on schedule. If hyperscalers slow or halt spending. That sounds simple, almost boring, until you trace what happens next. Every contractor who built out a data center on the strength of a hyperscaler contract, every chip supplier who borrowed against future orders, every engineering firm that leveraged itself to expansion — they’re suddenly holding a refinancing problem dressed up as an AI slowdown. The private credit vehicles funding all of this start repricing. The banks providing funding lines to those vehicles take the hit next. And because so much of this sits in leveraged, funding-dependent non-bank institutions, the BIS specifically warned that shocks can amplify through repo and derivatives markets. That’s the exact plumbing, by the way, that seized up in 2008.

I’ll add the wrinkle the report makes explicit and that deserves attention: this isn’t 2008 with new wallpaper. In 2008, the Fed had room to maneuver–rates had somewhere to fall, inflation wasn’t a constraint, and the federal balance sheet had runway. Today, any AI-credit stress would layer on top of inflation that’s still sticky, an energy market still digesting a Middle East shock, and sovereign debt levels that leave governments little fiscal cushion. The BIS’s own language for the moment: the global economy is caught in the “crosscurrents of progress and peril.” I don’t think central bankers reach for poetic phrasing like that by accident–that’s usually my bailiwick (🤣).

So what does the smart investor–the one sitting up in the cheap seats with the wide view–actually do with all of this?

Not panic. I want to be crystal clear about that, because it would be easy to read everything above and start drafting your hide-the-cash-under-the-mattress plan. That’s not the lesson here, and it’s not what the BIS is telling you either. The report explicitly notes that immediate risks to financial stability still look modest right now–this is a structure to monitor, not a fire alarm to pull. The AI buildout is real. The productivity gains are real. The capital being deployed is, for the most part, funding actual infrastructure that actual companies will actually use. None of that changes because a few rings of the circus are operating in dimmer light than the rest.

What changes is what you watch, and how closely. Keep an eye on hyperscaler free cash flow versus capex guidance on every earnings call from here forward (earnings season starts in a few weeks 😟)--the moment that guidance gets trimmed, the entire chain underneath it reprices at once, and you want to see that coming rather than reading about it after the fact. Watch credit default swap spreads on the major hyperscalers, which the BIS itself noted have already begun drifting wider. Watch the pace of private credit fundraising flowing into data center deals, because that capital is the refinancing lifeline for everything sitting off balance sheet, and lifelines have a way of disappearing exactly when you need them most. And watch the language coming out of institutions like the BIS itself. When the central bank for central banks draws explicit historical parallels, and this report does, to the dot-com bust, to 1840s British railway mania, to the Roaring Twenties, it isn’t doing so for color commentary. Every one of those episodes involved a real, transformative technology. Every one of them overbuilt. And in every one of them, it was the credit unwind that did far more damage than the equity correction that grabbed the headlines.

That’s the lesson the circus has been trying to teach me since I was a kid craning my neck to find whichever ring deserved my attention next: the spectacle that’s lit up isn’t always the one that matters. Sometimes the thing worth watching is happening quietly, off to the side, while everyone else is staring at the flaming hoop. The AI investment boom isn’t fake, and it isn’t doomed, but it is, increasingly, a credit story wearing a technology costume, and credit stories have their own rules, their own warning signs, and their own timeline that has nothing to do with how exciting the underlying technology happens to be.

Keep your eyes wide. Watch all three rings. And don’t let anyone–including me-convince you this is as simple as it looks from the cheap seats.

YESTERDAY’S MARKETS

Stocks rallied yesterday, with the Dow Jones Industrial Average closing above 52,000 for the first time as Alphabet's debut in the index helped lead blue-chip gains, while the S&P 500 added 1.17% and the Nasdaq Composite jumped 2.07% on a tech-led relief rally. The 10-year Treasury yield held around 4.38%, little changed after last week's sharp decline. WTI crude rose roughly 1.9% to settle near $70.57 a barrel.

NEXT UP

• Conference Board Consumer Confidence (June) may have improved to 94.4 from 93.1.

• JOLTS Job Openings (May) probably shrunk to 7.296 million from 7.618 million.

• Important earnings today: Nike and Constellation Brands