Dear Valued Clients and Friends,
In this week’s Dividend Cafe, I want to do something that I believe really needs to be done – explain what all the fuss is about regarding credit risk in the AI investment space.
“Ummmm, David, could you have possibly picked a more boring topic?” – you
“Yes, I could talk about the midterm elections.” – me [it will be less boring AFTER the midterms, so we’ll do that next week]
But this is not a boring topic. It is not merely about credit default swaps and counterparty risk and special purpose vehicles. And even if it were, that would merely be extra technical vocabulary applied to a topic that is very important and is not boring. But rather than say I want to talk about “credit risk in the AI space” today, let me put it this way … I want to talk about a topic that affects every single person, playing out right now across our economy, whether they have a dollar invested in AI or not.
I don’t do melodrama, and I certainly am no perma-bear (“how do you tell the difference between a perma-bear and someone who isn’t a permabear? A permabear never has any money“). I don’t do clickbait, and I don’t do shock-and-awe. If it would help my traffic to hype up this piece by trying to draw parallels between current AI financing/credit risk and the 2008 financial crisis, I would merely say that (a) I don’t care about my traffic, and (b) those comparisons are not offered in good faith – they are offered to sensationalize something. So what I am about to say is NOT analogizing 2008 and the AI moment because I can offer a thousand differences (but only need one – the balance sheet of the banks – and that’s all you need to know). But what I will say is that since 2008 I vowed to never, ever, ever underestimate credit risk again – and I have kept that promise. We have not had a systemic event like that the last 18 years (or anything close to it), but when certain credit dynamics caused me to raise an eyebrow over the years (2011 Europe, 2018 high yield, 2020 energy, 2026 private credit) I have diligently analyzed and scrutinized where warranted. Major market events almost never start in the stock market. They are almost always (almost) credit events that bleed. And so if this is boring, it is only because all of finance and markets and the economy must be boring to you, and if that is the case, why would you be reading Dividend Cafe? =)
So with that said, let’s jump into the Dividend Cafe …
| Subscribe on |
No Need to Read This Chart
I will start with a chart that I do not want you to read (or at least I am not asking you to). I am not sharing it because of any particular content or message in it. It is the visual, the complexity, the interconnectedness that you can intuit from the chart that I care about. It comes from Michael Burry, famously played by Christian Bale in the entertaining (if not cartoonish) movie The Big Short, who has been wrong about plenty over the years, but who is a robust and intellectually engaging read. I read him diligently (I read many, many people, and pay big subscriber bucks in aggregate, who I do not always agree with; if your advisor only reads those he agrees with, find a new advisor). All Michael is trying to do here is use one company to illustrate the vast circularity and interconnectedness within the AI ecosystem. It is wordy, clunky, and not graphically designed by the TBG graphic design team (to say the least). But I don’t think it leaves a lot of doubt around his underlying point.
*Cassandra Unchained Substack, Michael Burry, October 2026 1
When we talk about the credit and financing risk inside AI, we are nowhere near the point of merely analyzing one company and its order flow versus expense commitments (present and future). One company, recently thought of as the “second place” AI frontier lab, has over $537 billion of commitments for compute alone, another couple hundred billion dollars of commitments for TPU and GPU capacity – sometimes to companies who have a lot of cash flow (Google) and other times to companies that have none (SpaceX). But we are talking about an unprecedented level of spending commitment from a company with a tiny, tiny, tiny fraction of the revenues needed to cover such (and if you want to assume all of the revenue growth assumptions they, themselves, assume, all you get to do is take away a couple of the “tiny’s”) …
So then we start getting convertible notes here, preferred equity there, and all sorts of other balance sheet realities that were created by … guess who? … financial deals with their customers. Now, in fairness, a lot of this gets cleaned up in an IPO, hypothetically, as a lot of notes and prefs convert into various shares of common stock. And while a cleaner and more flattened balance sheet sounds like a good thing, we still have to deal with the fact that we have commitments to pay for things being made by a company with no money, to companies who are giving them things, lending them money, then converting it into equity – an equity that has value because of the presumed orders at the beginning of the sentence.
The Byzantine nature of it all makes me think Michael Burry’s chart at the top of this section deserves some credit.
Are People Getting Scared?
The cost of insuring against a default in Oracle’s debt has skyrocketed in recent months. This suggests growing concern about Oracle’s credit risk (see more in the next section).
But the volume of protection being bought on all credit default swaps has gone up 5x (500%!!!) in three months. Not only are investors paying more for it, but they are buying more of it, where the “it” is protection against default of debt from companies such as Oracle, SpaceX, Google, Amazon, Meta, Broadcom, and Nvidia. Total corporate CDS volume across the universe is much higher in nine months of 2026 than it was all of 2025.
But is this a sign of fear and loathing in Las Vegas (see what I did there), or is it just math? AI-affiliated debt issuance was about $100 billion last year – which now feels like someone’s lunch money. It has surpassed $466 billion so far this year, with plenty of time still on the clock. It stands to reason that a greater volume of debt issuance rationalizes a greater volume of insurance against that debt.
But I do not buy the idea that the cost of buying protection on Oracle’s debt going up (chart above) or even Meta’s and Nvidia’s (also higher) is a pure by-product of “fear of default.” I think it is worse than that.
I am not here to write that there will be no defaults in any of the bonds of any of these issuers, but if you put a gun to my head, I would not predict that these senior debt positions will face default (which is an entirely different category than what could happen to their equity). And yet, the increase in default protection is entirely rational as a by-product of liquidity in the marketplace, needed hedging costs, and market technicals. Meta and Nvidia both saw their CDS spreads widen over 100 basis points, and I don’t think a single buyer of that protection believes they are ever, ever, ever going to “default.”
Rather, they need the liquidity to buy more. And buying more means taking down the exposure on your books by buying protection. Rinse and repeat.
And yes, I have seen this movie before.
When a Simple Yes or No Won’t Do
Oracle has $273 billion of purchase obligations and uncommenced leases on its books, an earnings base that can not come close to covering this, $120 billion of other long-term debt, and has a wide array of accounting things going on that, if I got into it, would:
- Break my promise to not be boring,
- Sound really sinister when really they are legal, and
- Force you to ask, “Why are they doing that?”
And I think that last one is legitimate. Extending the useful life of assets to spread costs over a longer period of time is legal, but is the motivation concerning? Seeing revenues exceed what will really come in by 18-20% because of the way interest expense on customer pre-paid orders is treated is legal, but is it distorting what we really care about? My point is not to accuse anyone of anything untoward. My point is that there is a hornet’s nest of accounting complexities (best case), shenanigans (worst case), and potential can-kicking measures (my base case) that should cause us to ask, “why?”
That is where I have to be a skeptic, if not an outright cynic. I am paid to be paranoid and protective. And I am for companies doing all they can in the confines of the law and ethics to compete, improve market share, and grow. But if an extraordinary amount of effort, complexity, and even risk is going into something that creates zero new customers, zero new growth, and zero new market share – but rather is just intended to make it look better – am I allowed to ask, “why?” Do companies living in the golden age of something need to do that? Would they even want to? It seems fair to wonder.
The News Story or the News Reporter?
What I believe we are going to see in the next year or two is a real tension around whether or not credit conditions in AI are the news itself, or rather a sign of the underlying news … When financial company CDS spreads blew out in the fall of 2008, they were not merely telling markets, “hey, things have gotten bad at those Wall Street banks!” I am pretty sure markets knew that in the aftermath of Bear, Lehman, and AIG. But Morgan Stanley and even Goldman Sachs seeing their credit default swaps blow out became not merely a sign of the news, but the news itself – leading to a run on those banks, a decline in equity, and the need for significant new capital (thank you, Mitsubishi). The CDS story in AI thus far is child’s play compared to that, even for a truly troubled company like CoreWeave.
Higher credit default swap costs mean more collateral, tighter liquidity, a higher cost of capital, a greater urgency for equity deals, more restrictive commercial contracts, and at some level, a re-priced sector. The credit risk embedded in all of this is one aspect of it, but the need for finite resources to fund what feels like infinite need for expenditure forces more need for supply of protection, which enhances visibility, transparency, market signaling, and a reaction function around all of that.
Stock prices will be impacted – and the volatility of these stock prices will be impacted. The cost of credit and cost of protecting that credit will tell us about the AI story, and it will end up shaping the AI story. Mark my words.
Underneath it All
We not only have to think about the interconnectedness of all this in the context of the cost of capital (this is a capex boom, and nothing matters right now like capital), but questions about the return on this invested capital, the timing of those alleged returns, the ability to refinance (many, many loans are going to require extension), and how all of this impacts equity valuations. The risk with all of this underneath the surface is the classic negative feedback loop whereby a rising cost of capital limits profits on projects or kills some projects altogether, which reduces capital expenditures, which reduces revenues and profits for the major suppliers who are financing so much of this, which cuts off or limits some of the financing, which then increases cost of capital, which then …
Rinse and repeat.
And I am sorry, but if anyone thinks some version of that negative feedback loop is unlikely to happen, I’ve got a condo in Beaumont to sell you financed by New Century Financial in 2006.
Closing Caveats and Takeaways
We can talk about the valuations of many of the companies involved in the AI story. I think I have done that plenty. We can talk about whether or not the capex boom continues to uphold the “pick and shovel” companies who are providing the compute running this whole thing – and whether or not the growth rate of the capex boom can sustain those valuations. I have some opinions there. We can talk about the political and cultural mood around all of this, and how much of the story is vulnerable to those tides which may very well prove problematic. We can talk about whether the anticipated returns will exceed or fall short of expectations. All of these things are worthwhile conversations.
And of course, we can talk about where enhanced productivity may materialize, and how AI-generated efficiencies may provide a needed boost to the American economy. There is a partially full part of this glass, too; do not forget.
But from my vantage point, outside the vulnerabilities in the AI space is one glaring vulnerability that exists inside, outside, and within it – the credit risk that serves as a source of tremendous uncertainty for the sector, and for the broader market. A default in a neo-cloud company is not necessarily going to be a systemic event. A non-default spike in CDS costs for a hyper-scaler could be. Rising credit fears that impact the cash flow of already cash-limited operations are a razor’s edge. I do think some of this requires that Michael Burry chart-level of messiness to illustrate how interconnected so much of this is. And more than any capex boom in history, and more than any market moment I have ever participated in, this particular situation has very little margin for error. How could the vulnerabilities be any more obvious? The largest need for capital in history, all at once, is susceptible to more onerous terms on that capital and a lack of access to that capital, all without any clarity even in optimal conditions on the returns expected from this capex boom.
We intend to look closely at where credit markets indicate particular problems (neo-clouds) versus general problems (re-pricing across the space). And we do not intend to assume that it will all stay isolated if financing structures or, worse, a more fundamental questioning of the whole cycle comes to be. Credit spreads matter. CDS spreads matter. And the feedback loop can be monumental. It does not have to be 2008 monumental to be monumental. So to that end, we work.
Quote of the Week
“It’s not what you look at that matters. It’s what you fail to see.”
~ Lloyd Blankfein
More to Chew on
I am looking forward to being on the other side of the midterm elections on Tuesday, and that really has nothing to do with market vulnerability or positioning. I am just sick of the ads. I mean, really, seriously, totally sick of the ads. But I will do an ad-free discussion of the midterm results, two days after they are known to us, next Thursday (and it will be recorded).
In the meantime, have a wonderful weekend, set those clocks back Saturday, and bring on November. It’s a wonderful time of the year.
With regards,
David L. Bahnsen
Managing Partner, CIO
This week’s Dividend Cafe features research from S&P, Baird, Barclays, Goldman Sachs, and the IRN research platform of FactSet
(1) Third-Party Illustration — For Educational and Illustrative Purposes Only: Source: Cassandra Unchained, October 2026. Reproduced for commentary and illustration of the complexity of financial relationships surrounding Anthropic and AI infrastructure. The analysis and conclusions are those of the original author and have not been independently verified or endorsed by The Bahnsen Group or Hightower Advisors.
