Money Stuff: The AI Whateverpocalypse Trade

Okay so I have a business plan for Sam Altman. Here’s what you do. You build an artificial intelligence application for some specific sort of business, coding or tax planning or whatever. You roll it out on a Tuesday morning. You announce it in blandly optimistic terms and do some cool demos. You go on financial television, and when they ask you — as they will — a question like “won’t this disrupt the business model of software-as-a-service companies” or “won’t this put financial advisers out of business,” you smirk ruefully and shake your head and say “well there is a lot of disruption coming, and it may be difficult to know what role money will play in a post-artificial general intelligence world.” 

And by Tuesday afternoon every software company or wealth management company or whatever sector you’re targeting will have sold off by like 10%. Oh and of course the other thing you do is, on Monday — before you announce the new product — you buy a bunch of put options on all of the software/wealth management/whatever companies’ stocks. And you make like a 1,000% one-day return on your money, probably not in huge size but in useful size.

And then you do it again next week. You have, let us hypothesize, built something that is artificial, pretty intelligent and pretty general. You have a quite smart robot. “Robot, disrupt the X industry,” you tell it each week for some new X. And each week you eat some industry’s lunch and turn it into trading profits for yourself. Not investing advice!

Actually we have talked about this business model a few times before. I proposed it for OpenAI last year, for instance, and I also wrote about it when DeepSeek came out with a model that briefly crashed all the other AI stocks. And we’ve talked about the idea — “short an incumbent industry, and then disrupt it” — in other contexts; Joe Weisenthal proposed it decades ago.

Of course it is not specific to Sam Altman. OpenAI is a big AI lab, and I sort of like the idea of him smirking through this short thesis on TV, but any AI lab could do it, or they could take turns. Or, more generally, every AI lab could short every industry, and then they could disrupt each industry when they get around to it.

I feel like one way to express the model is:

1. Right now, there is an economy. People spend trillions of dollars on goods and services, and many of those trillions of dollars go to big public companies as revenue.
2. In N years, AI will satisfy all human needs and desires, so people will not spend any more money on goods or services, and all of those companies’ revenues will go to zero. As I once put it, in talking about DeepSeek: “Everyone has a magic lamp in their pocket with a genie that can grant unlimited wishes; the lamp is cheap and each wish is free.”
3. One possibility is that, if you control the robots, then in N years all of those trillions of dollars of revenue that other companies currently get will go to you instead.
4. But this is not the only possibility. Maybe robot control will be competitive, the genie will be nearly free, and all of the revenue will just vanish, just become consumer surplus. (Or perhaps all of the profits will end up with the chip makers, or the electric utilities, or something, but not the AI labs.) 
5. Or maybe the robot-controller will get rich, but it won’t be OpenAI (or whichever other lab). If you are building the robots, you want to hedge your risk of losing to better robots in a winner-take-all competition.
6. Also you don’t know what N is, and it might be a long time, and you need billions of dollars now to build more robots.
7. Also maybe this is all stupid. “Everyone has a magic lamp in their pocket with a genie that can grant unlimited wishes,” come on. You might want to hedge your risk that this is all wrong, that the revenue won’t vanish, that the industries won’t be disrupted, etc.

If the market now is both optimistic about AI and jumpy about its disruptive effect on other industries, you can monetize that jumpiness. If the market thinks that AI disruption will be this total, that AI will eat software and wealth management and everything else — or if the market just ascribes a meaningful probability to that — then you might as well capture that now.

The magic of financial markets is that they can turn a bad future fact about the world like “the revenue of Industry X will go down by $1 trillion” into a positive amount of money, for you, now. (A negative amount for someone else, sure, but that’s not your problem.) Also they can turn some probability of a bad future fact, or even some widespread beliefs about the probability of a bad future fact, into money for you now. Anyway:

On Wall Street, rising fears about artificial intelligence keep pummeling shares of companies at risk of being caught on the wrong side of it all, from small software makers to big wealth-management firms.

The latest selloff was triggered by a tax-strategy tool rolled out by a little-known startup, Altruist Corp. The perceived threat the sent shares of Charles Schwab Corp., Raymond James Financial Inc. and LPL Financial Holdings Inc. down by 7% or more, before going on to hit European companies, too.

It was the deepest slide for some of those stocks since the market’s trade-war meltdown in April. But it was only the most recent example of a sell-first, ask-questions-later mentality that has rapidly taken hold as new products emerge from the hundreds of billions of dollars poured into AI, sowing anxiety about how the technology may upend entire industries.

“Every company with any sort of potential disruption risk is getting sold indiscriminately,” said John Belton, a money manager at Gabelli Funds.

If AI is destroying the current economy to build something new, maybe make some money on the destruction first.

Companies often borrow money by issuing bonds. This gets them money, but it limits their flexibility. For one thing, they have to pay back the bonds: They have to make scheduled interest payments and pay back the principal at maturity, or else they will be in default and the bondholders will be able to sue to get all their money back. But that’s not all. Bonds might also have covenants saying that the company can’t do certain transactions — new borrowing, dividends, stock buybacks, etc. — and will often have change-of-control provisions saying that, if the company is acquired, it has to pay back its bonds immediately at 100 or, often, 101 cents on the dollar. The bondholders don’t want the company to do risky things — borrowing more, getting bought in a leveraged buyout, etc. — that would endanger their payments, so they negotiate covenants to prevent that.

Sometimes a company’s circumstances will change, and it will no longer want those limits on its flexibility. How can it get rid of them? Well, it could get rid of the bonds. How can it do that? Most corporate bonds are not freely callable; the company can’t just give bondholders their money back anytime it wants. It can call up each bondholder and say “hey we’d like to buy back your bonds” and negotiate a price, but that’s administratively annoying and probably won’t get everyone. It can do a tender offer, a public offer to buy back all of the bonds at the same price, but that probably won’t get everyone either. (In particular, if bondholders know that the company needs to get rid of the bonds, they might hold out for a higher price.) 

There is another, weirder way, at least sometimes. This is “defeasance.” Some bonds will have a provision saying that the company is allowed to defease them. To defease a bond, a company deposits all of the future interest and principal payments on the bond with the bond’s trustee, and then it washes its hands of the bond. Roughly speaking, the bond is no longer the company’s problem: The trustee has all the money it needs to make all the payments, so the company is off the hook. All those covenants — about not borrowing more money or doing mergers — no longer apply, because the bondholders no longer need to worry about the company’s credit, because the money is already there.

There is one important simplification in the previous paragraph: I said that the “company deposits all of the future interest and principal payments” with the trustee, but it normally doesn’t do that in cash. It normally does it in US Treasury securities. Consider, just hypothetically, a company that a few years ago issued a $750 million 30-year bond that matures in February 2051 and pays 2.95% annual interest. The company has to pay about $22.1 million of interest every year for the next 25 years (2.95% of $750 million), and then $750 million at maturity in 2051. If it decided to defease that bond, it could deposit $1.3 billion in cash ($750 million plus 25 years of $22.1 million interest payments) with the trustee, and then the trustee would have enough cash to pay off the bond.

But that would be dumb. There’s a US Treasury bond that matures around the same time and has a coupon of 3%. [1]  There’s about $54 billion of it outstanding. The company could just buy $750 million principal amount of that Treasury bond and deliver it to the trustee. Every year that Treasury bond would pay $22.5 million of interest, enough to cover the company’s bond’s interest payments, and in 2051 it would mature and pay $750 million, enough to cover the company’s bond’s principal. Perfect, done. You’ve replaced the company’s credit with US government credit, which seems like strictly an improvement. The bond is defeased, the bondholders are happy, and the company can walk away from the bond and its annoying covenants.

The nice thing about this trade, for the company, is that that Treasury bond is trading at about 74.5 cents on the dollar. The company could go out and buy $750 million principal amount of that bond for about $560 million. The company could completely satisfy its obligations under its own $750 million bond — a total of $1.3 billion of principal and interest — for a cash outlay of $560 million now.

That’s, you know, whatever, the magic of compound interest; the present value of those $1.3 billion of payments, discounted at the Treasury rate, is just $560 million. More specifically what is going on here is that both the company and the US Treasury issued 30-year bonds a few years ago, when long-term interest rates were very low, but now it is 2026 and long-term interest rates are higher. Back in 2021, a promise to pay $22 million a year for 30 years and then $750 million at the end was worth $750 million; now it is worth $560 million. [2]  That’s also the magic of interest rates.

Now forget about defeasance for the next two paragraphs. Say you bought that 30-year corporate bond back in 2021. You paid 100 cents on the dollar. Now interest rates are higher, and the bond is worth less. I said above 74.5 cents on the dollar, but that’s for a Treasury bond with no credit risk; a corporate bond with credit risk might be worth 65 cents on the dollar. Getting paid 2.95% interest for the next 25 years is no longer that attractive a proposition, so you have lost money. Oh well.

But you still have those covenants, and every so often they generate a windfall. For instance: What if the company has a change of control? What if there’s a leveraged buyout? Well, then it has to pay back its bonds at 101 cents on the dollar, right? Instead of paying you 100 cents on the dollar in 2051, which is worth about 65 cents today, the company would have to pay you 101 cents on the dollar right now, which is worth about 101 cents. That’s a big improvement!

Or it is, if you forgot about defeasance. But if the company can defease the bonds before its leveraged buyout, the covenants go away, and so does the change-of-control put at 101. Instead of getting 101 cents on the dollar today, you get exactly the contracted cash flows for the next 25 years, which are worth about 75 cents. [3] Oh well.

Anyway the company here is Electronic Arts Inc., which in 2021 did issue $750 million of 2.95% bonds due in 2051, along with $750 million of 1.85% bonds due in 2031, and which did announce a leveraged buyout in late September. (The LBO has not closed yet.) EA’s 2051 bonds were trading at about 65 cents on the dollar in September, before the deal was announced, but when it was announced they traded up to the 80s, and they were trading at around 92 cents on the dollar as recently as this Monday. Because those bonds do say [4] :

Upon the occurrence of a “change of control repurchase event” … each holder will have the right to require us to repurchase all or any part of that holder’s notes at a price equal to 101% of the aggregate principal amount of the notes.

So the LBO should trigger the holders’ right to get 101 cents on the dollar when the deal closes. But the bonds also say, further down and more boringly:

The indenture provides that we may discharge any and all of our obligations in respect of a series of debt securities, and, except with respect to certain provisions, [5]  the provisions of the indenture will no longer be in effect with respect to that series of debt securities on the 91st day after the date of the deposit with the trustee or paying agent, in trust, of money or U.S. Government Obligations in an amount sufficient to pay the principal, premium, if any, and interest on that series of debt securities, when due.

So one possibility seems to be that EA could defease the bonds (at about 75 cents on the dollar) instead of paying them off (at 101 cents on the dollar). (Not legal or investing advice! [6] ) 

Obviously 75 is less than 101, so EA (or rather its acquirers) chose that. Bloomberg’s Gowri Gurumurthy reports:

Electronic Arts Inc.’s bonds plunged Tuesday after the prospective buyers of the videogame maker launched a buyback offer that tied the notes’ prices to US Treasuries.

EA’s note due 2051 plunged 13.75 cents on the dollar to 78.25 cents on the dollar as of 2:39 p.m. in New York, according to Trace data. A bond that matures in 2031 fell 4.7 cents to 92 cents. Both are at their lowest level since the company’s buyout was announced in late September.

The slump came after EA’s acquiring group launched a tender offer for the $1.5 billion of notes that stopped short of offering investors face value, bucking the norm for such transactions. Instead, the bonds would be purchased at zero spread to US Treasuries, which weighed on prices.

Oops! People were expecting a tender at 101, and they’re getting a tender at 75-ish. Here is the tender offer announcement, which explicitly threatens:

To the extent any Notes remain outstanding following the consummation of the Tender Offers and the Consent Solicitations, the Company may (or the Offeror may cause the Company to) defease one or both series of Notes, in which case Holders of such Notes will continue to receive interest on each scheduled interest payment date and principal on the stated maturity date but will not benefit from any restrictive covenants removed pursuant to the defeasance, including the change of control repurchase obligations. 

Again, you do not see a ton of this; defeasance is a weird curiosity that does not come up a lot. (It’s not even in the index of Fabozzi.) But generally bondholders expect to get paid off at 101 in an LBO, especially if their bonds say “you will get paid off at 101 in an LBO,” so it is an unpleasant surprise for them to wake up and learn about defeasance.

In the long run, loans want to be securitized. When some new form of lending is invented — art loans or crypto loans or whatever — the company that invents it will probably both (1) make the loans and (2) own the loans. The company will have some money, and it will meet with potential borrowers and evaluate them, and if it decides to lend them money, it will lend them its own money. That just feels right. The lender knows the borrowers, it has decided to trust them, it understands this new form of lending that it just invented. People who didn’t invent this form of lending, who haven’t met with the borrowers, have no particular reason to give them money. The company making the decision is the company taking the risk. It has skin in the game.

But then the company will do this for a while and it will work pretty well and the company will make money and the new form of lending will take off, and eventually the company will realize that it has a lot of demand for loans and only so much money. So it will go to other people, outside investors, and say “hey we are making all these loans and they’re going great and earning a high return, do you want some?” And the outside investors will look at the company’s track record and say something like “hey you are earning high returns with low volatility and low correlation to other assets.” They will say this in part because it’s true — a newly invented form of lending should have high returns (no one else does it) and low correlations (it’s not like everything else) — and in part because there is limited data: A newly invented form of lending hasn’t been around for long enough for all of its risks to be realized.

But once it has been around for long enough to have some track record of good results, outside investors will be interested in getting in. So the company that makes the loans will be able to sell them to outside investors, who will not meet the borrowers and make the lending decisions themselves, but will trust some combination of (1) the pitch and documents and handshake that they get from the company that makes the loans and (2) the track record of the asset class. And ultimately there will be a division of labor between people who make the loans (who meet the customers and do the underwriting) and the people who own the loans (who put up the money). And this has various well-known problems, because the people making the loans have less skin in the game than they used to, though you can arrange the incentives to manage those problems.

This is all a sort of handwavy generic analysis and obviously not every sort of loan will be securitized, because not every category of loan will be big and popular and homogenous enough to be securitized. It is tempting to say something like “a private banker or auction house might lend you some money against your art collection, but that is a bespoke deal with one-of-a-kind collateral and would be hard to securitize to outside investors.” But in fact, sure, art loans get securitized. It’s a big enough category now.

Can Bitcoin margin loans be securitized? Oh absolutely. Lots of people want Bitcoin margin loans, and “Bitcoin margin loan” is a relatively straightforward concept for outside investors to understand and underwrite. If you write a bunch of loans against Bitcoin, and you have reasonably good methods for (1) taking collateral and (2) liquidating it when prices drop, then you can probably get outside investors to fund them.

It’s tricky to do it right this minute, though, what with falling Bitcoin prices and lots of liquidations. The Wall Street Journal reports:

Bankers at Jefferies have been pitching big investors for months on a $188 million bond sale backed by thousands of loans that crypto lender Ledn made to individuals. Ledn’s customers put up cryptocurrency as collateral for one-year loans, and the bond sale would give the company more money to lend.

But Ledn recently had to liquidate about one-quarter of the loans meant to back the deal, a person familiar with the transaction said. The forced sales happened after the price of bitcoin plunged by 27% since mid-January, triggering margin calls on the loans.

Asset-backed bonds—often associated with the 2008 financial crisis—are enjoying a renaissance as investors including big insurers and fund managers seek higher returns than conventional bonds offer. Ledn’s bonds are expected to pay 3 to 6 percentage points more than benchmark rates, according to data provider CreditFlow. ...

Jefferies originally told potential investors the Ledn bonds were to be backed by $199 million of bitcoin loans and $1 million of cash. The mix has since changed to about $150 million of loans and $50 million of cash, the person familiar with the transaction said.

The bond deal is still on track to close on Feb. 18, according to S&P Global Rating, which rated the bonds. Ledn must now extend new loans with the liquidation proceeds to generate interest income needed to pay bondholders. 

Oops. Ledn does margin calls at 70% loan-to-value ratios and liquidates at 80%; “so far, the mechanism has worked,” and “Ledn has liquidated 7,493 loans in its seven-year history at a maximum loan-to-value of 85% and has never experienced a loss.” It’s nice when a new asset class has a track record of earning high returns with low risk.

Thoma Bravo Inserts Lender Anti-Rebellion Clause In Buyout Loan. How Fake Invoices Duped BlackRock Unit Into a $400 Million Loan. ( Earlier.) Activist Investor Pushes Warner to Walk Away From Netflix Deal. Co-founders of Elon Musk’s xAI join exodus from start-up’s tech team. OpenAI Executive Who Opposed ‘Adult Mode’ Fired for Sexual Discrimination. OpenAI disbanded its mission alignment team. Quant Manager Qube Grows China Fund 10-Fold to Above $2 Billion. Elliott Takes Stake in London Stock Exchange Owner. Kraft Heinz Pauses Plan to Split Into Two Companies. Treasury Allows U.S. Companies to Provide Supplies to Venezuela for Oil Production. What It’s Like to Be a Banker in Russia During Wartime. Switzerland to vote on plan to cap population at 10mn. “Wall Street had a really good year, and that’s being reflected directly in Hamptons prices.” Goldman Sachs ups its late-night dinner budget for bankers to $35. What it was like to be a bush at Bad Bunny’s Super Bowl performance.

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks!