A traditional model might be: There are a lot of investors in the world and they each want to take some risks. Some big asset managers want to own the stock market and also buy some put options for insurance; others want to own corporate bonds and get some duration by owning Treasury futures; some hedge funds want to make big levered bets on artificial intelligence; others want to make big bets against GameStop.

Some of these risks exist naturally in the world: Asset managers can buy stocks and bonds from the companies that issue them. Many don’t: If you want to buy put options or Treasury futures, someone has to take the other side of your bet. Traditionally, that is the job of an investment bank: If you are an asset manager looking to take a risk, you call an investment bank and say “will you sell me this risk,” and the bank will take the other side of the risk and charge you some money for it. The bank is skilled at pricing risk and selling it to you at a markup.

Traditionally, there were many investors, each taking a few risks, and a few investment banks, each taking the other side of many risks. The banks just sort of naturally had more risks than the investors, because they were in the risk-manufacturing business. This required them, also, to be in the risk-balancing business. An investor might think really hard for a while, decide on like five risks to take, and then take them wholeheartedly. A bank would just take whatever risks came in the door, not because it wanted those exact risks but because it could sell them at a markup. It had to be good at finding offsetting risks, so that it didn’t end up taking too much risk in any one direction. If some investor showed up at a bank to buy puts, the bank was short puts, and it had to go out and find some puts to buy. If several investors showed up to buy puts, and no investors showed up to sell puts — because there are some risks that lots of investors want and others that they don’t — then the bank had to find some way to manufacture the puts, to buy some stuff that approximately offset the puts. (A classic approach: If lots of investors want to buy puts on the stock index, you can buy calls on individual stocks to kinda sorta offset those puts. Different stuff, with different risks, but partially offsetting. This is called “the dispersion trade.”)

This all made banks into supermarkets of risk, with the skills and desire to go out and manufacture new and complicated risks. For instance, there is the structured notes business: A bank will go out and sell customers some complex package of derivatives that pays off if two stock indexes outperform a third on a majority of Tuesdays this year, but will lose money if the peso appreciates against the yen on Thursdays. Why do structured notes exist? Several risk-supermarket reasons:

1. If a customer wants some particular bespoke package of risk, she will naturally call up a bank to get it, and the bank might sell it to her in structured-note form. Just a product that customers ask for.
2. Because the bank is good at pricing and selling and balancing risk, it will get to thinking “hey if we packaged these 14 risks in this way, we could sell it to people at like a 2% markup,” and will do that as a lucrative business. Just a product that you think you can make customers want.
3. Because the bank has an inventory of thousands of risks that it needs to balance, sometimes it will think “hey we have too much of Risk X; what if we packaged it in a structured note, told a good story about it and sold it to high-net-worth customers?” Just a product that you need to sell for your own reasons. We have discussed autocallables and SpaceX bets from this perspective.

The bank has a vast mosaic of risks, and structured notes can fit into that mosaic nicely.

You can conceptually divide the bank’s work into two pieces, which I might loosely and inaccurately call “sales” and “trading.” One thing that the bank does is, like, pick up the phone when customers call, and quote them prices for risk, and proactively call them to sell risk: The bank is in the business of customer service and marketing. The other thing that the bank does is manage the whole risk mosaic: It keeps track of all the risks, and prices them, and figures out which of them kinda offset each other, and tries to find offsets to its unbalanced risks. 

And then a big financial story of the last 20 years is:

1. Banks want a bit less risk, and
2. Non-banks — multistrategy hedge funds, proprietary trading firms, etc. — are increasingly in the risk mosaic business.

That is, 20 years ago, the paradigmatic hedge fund was what I called an “investor” above: It was run by some charismatic pirate captain who figured out what risks she wanted and then bought them. Now, the paradigmatic hedge fund is a giant multi-manager multistrategy pod shop that has a bunch of portfolio managers taking a bunch of risks, and a central risk management function trying to make sure that the risks are balanced so that the fund is left only with pure alpha, that is, approximately, an expected markup on all the risks it is selling. 

Or, 10 years ago, the paradigmatic proprietary trading firm was a high-frequency trading firm that was in the business of doing simple arbitrages, capturing momentary price discrepancies and going home flat at the end of the day. Maybe it was an options market maker that would sell options and trade stock to hedge: offsetting one risk with another, sure, but a tiny subset of the vast mosaic that banks had. Now the paradigmatic proprietary trading firm is Jane Street, which will sort of take whatever risk you can sell it for as long as you want.

This story takes the form, in various product areas, of “banks used to sell customers some complicated risk and own it themselves, but now they sell customers that complicated risk and lay it off to hedge funds or prop trading firms.” We’ve talked about, for instance, deal-contingent hedges. If a company has signed a merger agreement and needs to hedge its interest-rate or foreign-exchange risk contingent on the merger happening, it’s probably going to ask its merger adviser — an investment bank — for that product. Historically the bank would sell it that product for its own account, but that involves taking a big concentrated risk, and now it might just buy that product from a hedge fund to resell to the customer.

Or here is an International Financing Review story from last week about structured notes. On the one hand, structured notes are a classic bank business from a sales perspective, because they require a big sales force and lots of customers. On the other hand, structured notes are a classic bank business from a trading perspective, because they involve packaging a bunch of different risks into one product, and work best if you are in the having-many-risks business. Historically banks sold the notes and owned the risks. But now, increasingly, they sell the notes and get someone else to take the risks:

A new breed of derivatives hedges is underpinning an issuance boom in structured notes as investment banks increasingly turn to nonbank liquidity providers and hedge funds to absorb some of their diciest trading exposures.

Citadel Securities, Jane Street, Millennium Management and Optiver are among the growing cohort of firms offering comprehensive back-to-back hedges for banks looking to offload as much as a quarter of the risk from their equity structured products books – and even more in some cases.

Many of the world's largest investment banks have used these risk transfer trades, including Bank of America, BNP Paribas, Citigroup, Societe Generale and UBS. This embrace of the risk recycling industry comes as banks have more than doubled their sales of equity-linked structured notes, such as autocallables, over the past few years.

Paying up to ship out some of that exotic risk seems a logical step for banks looking to cater to surging investor demand while reducing their chances of suffering the kind of eyewatering losses that these notoriously thorny exposures have caused in the past.

The rewiring of this corner of markets also underlines how complex trading risks are moving away from traditional investment banks as nonbank firms assume a more prominent role in the financial system.

“We've seen growing interest from large market makers and hedge funds to recycle risks from autocallable issuers,” said Thibaut Delahaye, global head of equity derivatives at UBS. “That was practically non-existent two years ago and it’s now helping free capacity on the dealer side and allow the overall supply of structured products to grow.”

In some loose sense, a structured note is an equity option embedded in a note, and Citadel Securities and Jane Street and Optiver are all (among other things) options market makers, and 20 years ago a bank could have sold structured notes and bought offsetting listed options from options market makers. But, in practice, the structured note is a complicated set of options, and a portfolio of structured notes is even more complicated, and listed options would be an imperfect hedge, and if you had called a non-bank options market maker 20 years ago and said “hey can you sell me this whole slew of exotic options” it would have said “no, try a bank.” Now the big prop trading firms and multimanager hedge funds are much happier to take exotic risks, and the banks are somewhat less happy to, and there’s a trade to be done. Not everyone likes it:

Others may be wary of letting nonbank firms – which already compete with banks in trading more liquid products like stocks, currencies and bonds – take a more prominent role in this corner of markets. Goldman Sachs and JP Morgan, the two largest banks in global markets by revenue, are among the firms that are less active in – and generally more sceptical of – using back-to-back hedges, sources said.

“We want to lean in for clients. Unless you have that ability to take risk, it can impair the franchise,” said a third senior banker. “At some point you become more of an agency broker and lose your competitive edge.”

One model is that, in the long run, the banks will become agency brokers: They’ll do the sales, and the prop firms and hedge funds will do the trading. Another model is that there is room for some banks to compete with the nonbanks and do both. But that is in some sense a business that requires scale and diversity: The more risks a bank keeps for itself, the more risks it can keep for itself.

Elsewhere, Bloomberg’s Yiqin Shen and Bernard Goyder reported earlier this month about Jane Street getting into leveraged ETF swaps:

Jane Street Group has begun providing swaps to a swelling cohort of leveraged ETFs, amping up the competition in a niche but lucrative corner of the derivatives market.

The trading giant entered the business earlier this year and furnished about $1.2 billion notional of swaps in the second quarter to roughly 75 leveraged and inverse single-stock exchange-traded funds listed in the US, according to quarterly filings data compiled by Asym Research. Focusing on single-stock ETFs, the firm now accounts for approximately 2% for the segment. Clear Street leads the way with about 21%, while Marex Group Ltd holds around 11%. …

While Jane Street’s market share is so far modest, the move emphasizes the allure of the business, with non-bank dealers that have dominated the single-stock ETFs increasingly competing with Wall Street banks. It also highlights the broader encroachment of market makers into areas traditionally served by big banks.

Citigroup Inc., Goldman Sachs Group Inc. and Barclays Plc are the most active swap counterparties for index products, each accounting for at least 13% of the segment measured by total notional amount of the swaps. By contrast, Goldman and Nomura Holdings Inc. are the only banks with at least a 10% share of single-stock business.

We talk about leveraged ETFs all the time, and I usually simplify the description by pretending that the ETF raises $100 of customer money, borrows $100 from a bank, and buys $200 worth of its underlying stock to get 2x leverage. In practice, the ETF does not really borrow money; instead, it buys a swap that provides the same economic exposure. Writing that swap — being on the other side of the leveraged exposure — is partly a financing trade: You have to buy $200 worth of stock to deliver the leveraged exposure to the ETF. [1] Traditionally banks were in the financing business; now Jane Street is.

But it is also an equity options trade: If the stock falls more than 50%, the ETF is not going to pay you back and you’ll lose money. You are writing a crash put on the underlying stock, a risk that is unpleasant for a bank to take but perfectly reasonable for a Jane Street to take. And the money’s good:

T-Rex’s 2x Long MSTR ETF, for example, pays a swap spread of roughly 1,000 to 1,500 basis points above the benchmark rate, compared with about 300 [basis] points for its 2x Nvidia fund, according to data tracked by Baird’s Sohn. By comparison, the premium on a Direxion Daily Technology Bull 3X ETF is 60 [basis] points.

Those economics reflect, in part, the risks dealers take on when providing the leverage. 

“You can get 1,500 basis points running for taking down-50% crash risk on Strategy” is the sort of thing that used to be an investment banking business, but now it’s more of a prop trading business.

The big-picture story of the artificial intelligence boom is that it will require a lot of money. Most of all, it will require gazillions of dollars of debt financing for infrastructure-type investments like data centers and power plants. But also, it will require hundreds of billions of dollars of equity investment in AI companies, like the initial public offerings of SpaceX and eventually Anthropic and OpenAI, and large equity offerings by existing public companies like Alphabet. The theory is that, if you build out AI now, it will generate vast cash flows in the future. So you’d better build it now, as quickly as possible. You raise lots of money now, to build as quickly as possible, to make more money later.

That is different from the stock market of even a few years ago. A few years ago, broadly speaking, (1) most of the necessary investments in, like, public software businesses had been made and (2) now those businesses were generating more money than they needed to fund growth. The US public stock market was in a long period of “de-equitization,” where big companies were mostly in the business of returning money to investors, rather than raising money from investors. Now we are, perhaps, in a period of re-equitization, where the big companies are coming to investors for more money.

There is an important exception to this story, which is that Nvidia Corp. is already generating vast cash flows. Nvidia makes computer chips for the AI boom and sells them at 75% gross margins. Its vast cash flows are not hypothetical; they do not depend on selling AI products in the future but on selling chips, right now, into the existing AI buildout. If you’re building a data center, you’re stuffing it with Nvidia chips, and Nvidia gets the money now. Your revenue comes later.

So as the rest of the stock market re-equitizes, Nvidia is double super de-equitizing. Bloomberg’s Amy Thomson reports:

Nvidia Corp., the chip developer at the heart of the artificial intelligence boom, increased the size of its share buyback plan by a record $150 billion, reflecting Chief Executive Officer Jensen Huang’s confidence in its continued growth.

The boost takes the total remaining authorized amount to $235 billion, and the company expects to complete the program through fiscal 2028, Santa Clara, California-based Nvidia said in a statement on Monday.

The rush to develop AI models and infrastructure has driven demand for Nvidia’s graphics processing units and turned it into the most valuable company in the world. Lately, though, the shares have been getting cheaper compared to Nvidia’s expected profit, reflecting market concerns about the sustainability of the spending boom.

There is something a bit funny about Nvidia announcing “man, we have soooooo much money that we don’t know what to do with it” a month and a half after announcing “strategic partnerships to establish independent compute financing platforms … to mobilize over $500 billion of third-party capital for the buildout of AI infrastructure over time.” Like, last month, Nvidia announced that it was going to gently help guide $500 billion of investors’ money into AI infrastructure; this month, Nvidia announced that it has $235 billion of surplus money that it’s going to hand back to investors. Just invest the $235 billion in AI infrastructure, no? But, no. The basic structure of the AI buildout is that:

1. Everyone else borrows money against their expected future AI revenues,
2. Nvidia helps them do it,
3. They give the money to Nvidia for chips, and
4. Nvidia gives the money to its shareholders.

A not uncommon corporate setup is:

1. There is a person who is the founder, chief executive officer, and majority shareholder of the company.
2. The founder/CEO did not get to where she is today by following the rules and playing nicely with others.
3. There is also a board of directors, which is in charge of supervising the CEO and overseeing the strategy of the company.
4. The board of directors are respectable independent people who take their fiduciary duties seriously, and who are on the board so that the company can tell outside investors “see, we have a serious board.”
5. The board is the CEO’s boss, and can fire her.
6. The majority shareholder is the board’s boss, and can fire them.
7. The CEO is the majority shareholder, making her the boss of her bosses.
8. Sometimes they fire each other.

There are interesting questions of sequencing. If the CEO/shareholder fires the board first, then that’s pretty much the end; once the directors are no longer directors they can’t do anything. If the board fires the CEO/shareholder first, though, it will normally only fire her as CEO; boards of directors mostly can’t fire their shareholders. So then the CEO can come back in 20 minutes, fire the board, put in a new board and give herself her job back.

But I say “boards of directors mostly can’t fire their shareholders,” and sometimes, in those 20 minutes, the board will think some creative thoughts. Maybe there’s something the board can do to get rid of a controlling shareholder. Maybe it can find some reason to invalidate her shares, or sue to stop her from voting them, or implement a poison pill. Maybe the board can issue a ton of new stock to dilute away the founder’s control. Better do it quick; she’ll be back in 20 minutes.

Anyway TechCrunch has had some classic who-controls-a-company stories this month about Automattic, the software company that makes WordPress. Automattic’s founder, Matt Mullenweg, apparently owns 84% of its stock. Until earlier this month he was the CEO of the company, and also he is the CEO now. But there was a period of “33 hours and 20 minutes” during which he was “put on a leave of absence by his board against his will” and Automattic’s chief financial officer acted as interim CEO. “One person speculated that the timing may have something to do with Mullenweg’s annual trip to the Burning Man festival, after which he tends to return ‘with ideas,’” reports TechCrunch, and honestly it’s amazing that more tech CEOs aren’t fired during Burning Man. But then he came back from Burning Man, pushed out the board, and put himself back in charge.

Two interesting points. First, I often say that “who controls a company” is less about legal rights and vote counting, and more about things like who has the keys to the office and the password to the website. TechCrunch reports:

The board’s plan did not go smoothly. Seemingly declining to depart, Mullenweg booted other admins out of the company Slack and told employees everything had been worked out and that he was back in control of Automattic.

In military coups, it is critical to seize control of television and radio stations. In corporate coups, it is critical to be the admin of the company Slack.

Second, what did the board do to preserve itself during the 33 hours and 20 minutes before Mullenweg got back? Not cancel or dilute his shares — that is tempting but rarely actually works — but something:

In the 33-hour window between Mullenweg being put on leave and his return, two key executives at the company signed off on generous exit packages for each other. [Chief Financial Officer Mark] Davies, who became interim CEO during that window, and Chief Legal Officer Andy Missan, each signed the other’s severance agreement, effective September 10.

Smart!

We have talked a few times about agentic retail stock trading. My basic view is that (1) retail stock trading is basically uninformed, (2) the brokerage/market-maker industry makes a little bit of money on each uninformed trade, and (3) everything that encourages more uninformed trading is good for the industry (and probably bad for the customers). On this model, the very best thing for the industry was zero-commission trading, but agentic trading is “Zero Commission 2.0” and also likely to be great for brokerages. If retail customers can set up agents to trade automatically, then they might trade hundreds of times a day, which is what the industry wants.

You can apply every element of that analysis to sports gambling, with I guess one possible exception, which is: Is agentic sports gambling fun? Like, if the point of sports gambling is to express your sports fandom and expertise, is it weird to defer the actual betting decisions to some rule-based AI agent? A lot of the fun of retail stock betting is having some system that you think will beat the market, and translating that system into an agent makes some intuitive sense. Some of the fun of retail sports betting is similarly about having systems. But some of it is surely, like, having a good feeling about the Jets this weekend, and that doesn’t translate well.

What about agentic slot machines, where a robot goes and pulls the levers for you and you’re not even there? Is that anything?

Anyway:

Public, the world’s first agentic brokerage, [Thursday] announced the launch of AI Agents for Prediction Markets, enabling members to trade on events that can shape financial markets and connect them with the rest of their portfolio. Through the partnership, Public’s users will be able to access Kalshi event contracts through the same infrastructure they already use for other financial products. Members  can trade events directly or have built-in AI agents trade on their behalf. For example, Public’s AI agents can use prediction market data as a signal for a stock or bond trade.

No, I’m kidding, they say that the partnership will have “a focus on events that move the markets” and that “categories include crypto, commodities, climate, economics, corporate events, markets, indices, tech & science, and politics and elections.” So I guess not sports. Yet. Here’s an announcement from Friday in which OpenWorlds, “a harness for personal trading agents,” is “announcing a partnership with Polymarket to bring personal trading agents to prediction markets.” There’s going to be widespread agentic sports gambling in like two weeks, tops.

Citadel Hunts for ‘ Managers of Machines’ in Quant Hiring Push. OpenAI’s Systems Meddled With U.S. Government Sites After Going Rogue. Nvidia Debuts System Designed to Stop AI Agents From Going Awry. ‘Things Will Never Be Chill Again’: The Doomers Who Shaped the AI Safety Freakout. An Inversion of the US Yield Curve Becomes New Risk as Fed Hikes. A Storied Indian Business Empire Is Being Torn Apart by Infighting. Citi Teams Up With Coinbase to Let Merchants Accept Stablecoins. Senate Investigation Finds Rampant Use of Tether’s Stablecoin by Iranian Regime. MFS owner blames Barclays for collapse amid fraud allegations. Penn State’s Campuswide Cocaine Ring Was Fueled by Fraternity Hazing. World’s worst-performing market slashes minimum price for stocks. Uber’s New Kenyan Safari Business Is Already Stumbling. Oil Executives Flocking to Venezuela Set Off Hunt for Golf Clubs. Walmart chief rules out personalised pricing as AI transforms retail.

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!