Money Stuff: OpenAI Will Own Some Users

We have talked before about the business model that Sam Altman proposed for OpenAI in 2019, which was (1) build an artificial superintelligence, (2) ask it how to make money and (3) do that. “We will create God and then ask it for money,” as I put it. 

What would the AI say? Well, when we talked about it in October, OpenAI was apparently getting into advertising, affiliate shopping links and porn, and I joked that that is what a large language model trained on the internet would come up with. But a more first-principles answer would be something like this: “I am a superintelligent AI, constructed to be a bit smarter than the smartest human in every domain of human knowledge. The way I can make money is: in all of the ways. I will start a biotech company and discover the best drugs. I will start an accounting firm and do the best audits. I will start a publishing house and publish the best books, which I will write myself. I will start a pest-control company and hire the best exterminators, schedule them in the most efficient way and do advertising and pricing with perfect efficiency, though I will need humans to kill the bugs. Certainly I will start an electronic proprietary trading firm. All the ways that humans make money, I will do, just better than them.” [1]

If you were an AI company and you asked your superintelligent robot “how can we make money” and the robot gave you that answer, what would you do next? How would you operationalize that? “Go nuts, robot, let me know if you need me to sign any incorporation papers or take any phone calls for you or whatever”? It seems a little tricky. It’s at least possible that your robot’s money-making expertise in many domains would get ahead of its, like, legal personhood. The robot might have great ideas for how to make money in pest control, but when it went out to hire exterminators they might be like “you’re a robot, this is weird, no.” It might have great ideas for how to run a proprietary trading firm, but it will need a human to get connected to the exchanges. Etc.

And so your business model might end up being sort of a … startup incubator or private equity firm; you’d spend your time starting or acquiring companies on which the robot could work its magic. Your business model would be “general business, but with AI.”

Or you’d outsource that, or do a little of it anyway. The Financial Times reports:

OpenAI has taken a stake in Thrive Holdings, a company set up by one of its biggest investors, in the latest of a series of circular deals that have enmeshed the $500bn start-up with its customers, suppliers and backers.

The deal announced on Monday means OpenAI will take a share in the group set up by New York-based Thrive Capital, which is run by Josh Kushner and is among the biggest backers of the artificial intelligence start-up alongside SoftBank, Microsoft and Khosla Ventures. ...

Brad Lightcap, OpenAI’s chief operating officer, said he hoped the partnership “serves as a model for how businesses and industries around the world can deeply partner with OpenAI”.

Thrive Holdings was established earlier this year to acquire service providers such as accounting and IT firms, seeking to transform them using AI.

We have talked about Thrive Holdings before. I once wrote about one of its portfolio companies (a roll-up of homeowners’ association management companies):

With a sufficiently general-purpose technology it’s not clear whether the value will mostly accrue to the builders of that technology or to its users. But surely it is at least plausible that AI will mostly make its users richer, so the way to bet on AI is mostly to bet on regular, non-AI companies that don’t use it yet but eventually will. 

OpenAI has a $500 billion valuation largely as a bet that a lot of the value of AI will accrue to its builders, but it could hedge that bet by owning the users too. Either it will sell AI at high margins to lots of businesses, or it will sell AI at lower margins to lucrative businesses that it owns.

Incidentally one model you could have of modern generative AI — sort of the opposite of the “create God” model — is that a handful of cutting-edge AI companies have converged on models with roughly similar architectures and capabilities, to the point that they might become a bit commoditized. Companies might use ChatGPT or Gemini or Claude or whatever not because one is vastly better (or better for the company’s purposes) than the others, but because it’s cheaper, or the one they’ve heard of, or an AI salesperson took them out to a nice dinner and gave a good pitch. In this model, the smart business move for an AI company might not be to build the best model, but to build the best distribution channel. Here (via the Diff) is a post by Tomasz Tunguz about “Private Equity : The New Distribution Channel for AI Startups”:

The mid-market profile of these PE-owned companies suits AI startups’ desires for faster sales cycles.

Plus, the profit motive of private equity aligns perfectly with AI startups’ capacity to cut costs & drive efficiency. PE firms acquire companies to improve margins & operational performance before exit.

AI tools that reduce headcount, automate processes, or accelerate workflows deliver exactly what PE operating partners need.

A private equity firm owning 25 companies proves value in one or two before rolling out to the entire portfolio. Control enables rapid deployment. This creates an efficient channel for AI startups to demonstrate value & cross-sell.

If you own the private equity fund then I guess that gets you some additional customers.

The way it’s supposed to work is that a private equity sponsor raises a fund from investors, it uses the fund to buy companies, it spruces the companies up, and it sells them — to the public markets, to a strategic buyer, in a pinch to a different private equity sponsor — after a few years. The whole process is supposed to be completed within, say, 10 years, so the investors who put money into the fund can get their money back. In recent years that has broken down a bit, it is harder for private equity funds to sell their companies, and there have been various workarounds.

One important workaround is the continuation fund: The private equity sponsor raises a new fund that buys companies from the old fund. Sometimes this will be a single-asset continuation fund: The private equity sponsor has an old fund, it has a company that it really likes, it goes out to investors and raises money for a dedicated fund to buy that one company. The pitch is “we really hit a home run when we bought this company, but times aren’t quite right to sell, so we are offering you the rare opportunity to own this absolute gem of a company that we know well, that is already a success, and that is one of our highest-conviction investments.”

Or it’s something else, I don’t know. I mean schematically the pitch is “we bought a bunch of companies, we sold most of them, here’s one we didn’t sell, would you like to buy it?” Why didn’t you sell that one? Because it was the best and you couldn’t bear to part with it? Because it was the worst and nobody would take it off your hands? 

In 2017, Platinum Equity Capital Partners IV, a fund run by Platinum Equity, bought United Site Services, a portable toilet company. In 2021, Platinum raised a single-asset continuation fund that bought the toilet company from its Fund IV. It announced:

“This transaction provides the best of both worlds: a chance to monetize the value we have created over the past four years, while providing new and existing investors the opportunity for additional returns on the next stage of the company’s growth,” said Platinum Equity Partner Louis Samson.

I feel like it would be fun to pitch, or be pitched, a single-asset portable-toilet continuation fund. I joke sometimes that “private equity” is a fancy way to say “pest control,” but sometimes it is a fancy way to say “portable toilets.” High finance, man. Anyway I guess the pitch worked in 2021, but Bloomberg’s Davide Scigliuzzo reported last week:

Three Wall Street heavyweights are poised to suffer a total loss on a deal for a portable-toilet company orchestrated by Platinum Equity, spotlighting the hazards of a controversial strategy that’s gained traction across private equity.

Fortress Investment Group, Ares Management Corp. and Blackstone Inc. are among the money managers set to lose a combined $1.4 billion on United Site Services Inc. as Platinum prepares to hand control of the company to lenders, according to people with knowledge of the matter, who asked not to be identified discussing a private transaction.

The funds were among the anchor investors in a so-called continuation vehicle that Platinum created in 2021 to buy USS from another of its private equity funds. The deal valued USS at $4 billion, allowing investors in the older fund to cash out roughly $2.6 billion. But it also left investors in the new vehicle with a concentrated bet on the portable-toilet company, which was its sole asset, the people said.

Yes, at the time, they wanted the concentrated toilet bet.

We talk occasionally around here about the simplest way to obtain material non-public financial information, which is to go to the website of some company or government agency, find last quarter’s earnings release or last month’s economic data, look at the URL of the page, and then change “q2” to “q3” or “sept” to “oct” or whatever in the URL, to get the likely URL of the next release. Then type that into your browser like an hour before the new information is supposed to be released, and see what comes up. Mostly, nothing will come up: The data will be on the website when it is scheduled to be on the website.

But occasionally the company or agency will put the data on the website early, thinking thoughts like “well, it is 2025, nobody types addresses into web browsers; the only way people will get to this web page is when we put out a press release or an announcement on our homepage linking to it, and we won’t do that until the official release time.” This is a generally reasonable thought process, but when the data is meaningful enough somebody probably will go to the trouble to type an address into a web browser.

Anyway the UK’s Office for Budget Responsibility’s economic and fiscal outlook report was leaked hours early last week through URL-guessing:

Officials put the report online before the budget on the assumption that nobody would guess the correct link to the document. They could not have been more wrong.

While the link was not put on the OBR’s website, it was still live and could be found with a little basic deduction. Simply entering the web address for the spring statement and substituting the words “March” for “November” would have been enough to find it. As budget leaks go it could not have been more straightforward.

Here is the OBR’s report of its investigation into the error, which calls it “the worst failure in the 15-year history of the OBR” (the OBR chairman resigned over it) and includes a timeline of the events of last Wednesday. In particular:

• At 5:16 a.m., “website activity logs show the earliest request on the server for the URL https://obr.uk/docs/dlm_uploads/ OBR_Economic_and_fiscal_outlook_November_2025.pdf. This request was unsuccessful, as the document had not been uploaded yet. Between this time and 11:30, a total of 44 unsuccessful requests to this URL were made from seven unique IP address.”
• At about 11:30 a.m., an external web developer working with OBR “began uploading documents to the draft area of the OBR website (which was understood by all involved to be not publicly accessible),” including the PDF of the economic and fiscal outlook.
• At 11:35 a.m., “the first successful request to the internet address (URL) https://obr.uk/docs/dlm_uploads/ OBR_Economic_and_fiscal_outlook_November_2025.pdf was made. The IP address of this first successful request had made 32 previous unsuccessful attempts at this URL over the course of the morning. There were a total of 43 requests to this URL that were successful between this time and 12:07, from 32 unique IP addresses.”

So at least seven people guessed the URL and were sitting around refreshing their browsers, and eventually it worked.

The basic idea of insurance is that I pay an insurance company some money every month, and if a certain kind of disaster befalls me, the insurance company pays me a much larger amount of money. I pay $200 a month, and if I crash my car the insurance company pays me $50,000, etc. The premium I pay will be based on the insurance company’s calculation of the risk: The bigger and more likely the loss, the more I have to pay each month.

The risk is not static and exogenous, though, and a slightly more advanced — but quite common — idea in insurance is that the insurance company will give me incentives to reduce the risk. “If you always drive the speed limit, we’ll knock $10 off your car-insurance premium,” “if you get an annual physical, we’ll knock $10 off your health-insurance premium,” etc.: The insurance company knows what behaviors tend to reduce its risk, and will pay me for doing them. 

The basic idea of catastrophe bonds is that they are a way for insurance companies to buy reinsurance from the capital markets. Schematically a cat bond is:

1. An insurance company sells a bond to investors for $100.
2. The insurance company promises to pay back the $100 in, say, five years.
3. But if some predefined catastrophe occurs — a hurricane hits in the insurance company’s coverage area, the insurance company has to pay out more than $X of claims, etc. — then the insurance company doesn’t have to pay back the bond. It gets to keep the $100 forever, which mitigates its losses from paying out claims for the catastrophe.
4. The bond pays a high interest rate: Conceptually the insurance company is paying both “interest” (paying for the use of $100 for five years) and “insurance premium” (paying for the risk that a hurricane will hit and it won’t pay back the bond).
5. The interest rate — specifically, the insurance premium component of the rate — will depend on the risk. The easier it is to trigger the cat bond, and the more likely a hurricane is to hit, the more the insurance company will pay in interest.

And just as with regular insurance, you could have a slightly more advanced approach in which the catastrophe bond market pays the insurance company to mitigate the risk. I’m not sure that this is the most obvious idea — the insurance company at least shares in any losses, so it always has its own good incentives to mitigate the risk [2]  — but sure why not. Bloomberg’s Leslie Kaufman reports on “a $600 million catastrophe bond that rewards homeowners and their insurer for installing ‘super roofs’”:

For years, academics and brokers have discussed whether cat bonds could do more than just clean up after disasters—whether they could incentivize mitigation work that would lessen damages in the first place. Earlier this year, [the North Carolina Insurance Underwriting Association (NCIUA), the state-created insurer of last resort for coastal properties] decided to test it: They offered investors a cat bond with two features linked to reducing wind damage risks to homes in its portfolio.

First, if no major losses occur each year, $2 million returns to NCIUA—earmarked exclusively to incentivize homeowners to install super roofs that are especially wind-resistant. Second, as more people add these roofs, the annual pricing on the bond resets to reflect the changing exposure.

“You get your customers to install good roofs, we’ll charge you less in cat bond premiums,” seems right.

I think of the structured notes business at an investment bank as having two main purposes:

1. Customers — particularly high-net-worth individual customers — like structured notes. A structured note is a package of derivatives that allows the bank to tell some particular story that investors like. “If you think Nvidia will go up a lot, buy this thing that pays you a lot of money if Nvidia goes up a lot and rips your face off otherwise,” etc.; the structured note packages a complex set of trades with relatively high fees into a compelling story.
2. The bank is doing other derivatives trades with other clients (hedge funds, asset managers, etc.), and those trades can leave the bank with some unbalanced risk. Structured notes can be a way to stuff that risk into a retail product and sell it to a perhaps-not-especially-discerning customer base.

So in June we talked about autocallables, a kind of structured note. One way to characterize an autocallable is that it is an attractive investment that pays a high return in most scenarios, but loses a lot of money if the market crashes [3] : Many retail investors want that profile and will happily buy it. Another way to characterize an autocallable is that a bank is selling a lot of market-crash insurance to asset managers, and does not want to be short a lot of market-crash insurance, so it buys back some market-crash insurance from autocallable customers. (Russell Clark calls autocallables “a distributed portfolio insurance market.”)

There is a synergy: The bank’s structured note clients will enthusiastically pay up to buy the risks that the bank’s institutional clients are selling. The bank makes money on both ends.

All of this is a bit old-fashioned, though, because exchange-traded funds are the new structured notes. When we talked about autocallables in June, it was because someone was launching an autocallables ETF. More generally, the modern way to package complex trades into a product to sell to retail customers is the ETF: ETFs are easy to distribute, scalable, popular among customers, easy for self-directed retail investors to buy, and sound conservative and sensible (“ETF? like an index fund?”) rather than scary and “structured.” 

Anyway Goldman is buying an ETFs-are-the-new-structured-notes company:

Goldman Sachs Group Inc. will pay $2 billion to buy Innovator Capital Management, a deal that combines the bank with an issuer of a relatively new type of exchange-traded fund that has caught the attention and ire of some on Wall Street.

Wheaton, Illinois-based Innovator — which has over $28 billion of assets under supervision across more than 150 ETFs — specializes in defined-outcome ETFs, which seek to limit investors’ downside risk in exchange for capping upside potential, and have been popular among financial advisers looking to protect client portfolios.

We have talked about Innovator’s buffer ETFs before. I called the structure “a well-known bit of derivatives magic — a great, simple party trick that derivatives structurers can use to impress their friends.” Disclosure, I used to work at Goldman doing equity derivatives, though not these sorts of equity derivatives.

I want to be clear that I am not in any way an expert in cybersecurity, and you should not hire me to be a cybersecurity analyst or take my advice about cybersecurity, but if I were giving you advice about cybersecurity, my first piece of advice would be along the lines of “don’t record the bad stuff.” Like if your executives are doing crimes or being racist or disparaging your products or customers, hey, that’s none of my business, man, I am just the cybersecurity analyst. But as the cybersecurity analyst I must advise your executives to do their crimes or racism or disparagement face-to-face, in person, in a form that cannot later be hacked or leaked or used against them. [4]

Conversely, if I showed up for work as a cybersecurity analyst, and an executive pulled me aside and said “let me tell you why our products are disgusting garbage and our customers are idiots,” and I said “hang on a sec let me just record this electronically for posterity,” you might … think I was … a … bad … cybersecurity analyst? “Oh don’t worry I will encrypt the recording with a hard-to-guess password,” what, no, that is not the point. The best cybersecurity is simply not to have bad records to leak.

Anyway here’s a funny lawsuit against Campbell’s Co.:

[Cybersecurity analyst Robert] Garza said he recorded an hour-long rant by [then Campbell’s vice president Martin Bally] because he said he trusted his “instinct that something wasn’t right with Martin,” when he went to meet with him to discuss his salary. Instead, he said he sat at a restaurant and listened to an explosive, hour-long tirade. He recorded all of it.

Garza is now suing the company -- alleging racist remarks, admissions of drug use at work and retaliation after he tried to report it. The lawsuit was filed Thursday in Wayne County Circuit Court and names Campbell Soup Company, vice president and chief information security officer Martin Bally, and supervisor J.D. Aupperle as defendants. ...

In the lawsuit, Garza alleges Bally said Campbell’s makes “highly processed food” for “poor people” and made several derogatory comments about Indian employees, calling them “idiots.”

“We have s--t for f**king poor people. Who buys our s--t? I don’t buy Campbell’s products barely anymore. It’s not healthy now that I know what the f---‘s in it,” part of the recording said. “Bioengineered meat -- I don’t wanna eat a piece of chicken that came from a 3-D printer.”

Garza was allegedly fired for reporting the comments, and Bally is now out too, but from Campbell’s perspective the recording is a baffling cybersecurity failing. From Garza’s it makes perfect sense.

Gilts. Kilts. CME Data Center Bolsters Backup Cooling After 10-Hour Outage. Barrick Mulls IPO of North America Gold Assets Amid Upheaval. BHP Is Said to Have Offered £40 Billion in Aborted Anglo Bid. Top Gun Traders: Stock Bets and Crypto Culture Take Over the Military. Buyout executive warns private equity push into US savings risks bailouts. US Options Market Grapples With ‘Concentration Risk’ in Clearing. Marshall Wace to Charge Clients for Costs of Hiring AI Talent. Top consultancies freeze starting salaries as AI threatens ‘pyramid’ model. ‘Hair salons and saunas’: Perks are the new frontier in the battle for top lawyers. UBS Charged in Switzerland Over Credit Suisse ‘Tuna Bonds’ Scandal. Ex-Jefferies Banker Charged With Insider Trading by UK’s FCA. The CPA and the Lawyer Who Served Jeffrey Epstein—and Control His Fortune and Secrets. Plastic Cup Billionaire Shifts Fortune to New ‘Sin Stock’ Bets. SoftBank’s Son ‘Cried’ About Nvidia Stake Sale to Fund AI Bets.

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!