| There’s a famous Sam Altman interview from 2019 in which he explained OpenAI’s revenue model [1] : The honest answer is we have no idea. We have never made any revenue. We have no current plans to make revenue. We have no idea how we may one day generate revenue. We have made a soft promise to investors that once we’ve built this sort of generally intelligent system, basically, we will ask it to figure out a way to generate an investment return for you. [audience laughter] It sounds like an episode of Silicon Valley, it really does, I get it. You can laugh, it’s all right. But it is what I actually believe is going to happen.
It really is the greatest business plan in the history of capitalism: “We will create God and then ask it for money.” Perfect in its simplicity. As a connoisseur of financial shenanigans, I of course have my own hopes for what the artificial superintelligence will come up with. “I know what every stock price will be tomorrow, so let’s get to day-trading,” would be a good one. “I can tell people what stocks to buy, so let’s get to pump-and-dumping.” “I can destroy any company, so let’s get to short selling.” “I know what every corporate executive is thinking about, so let’s get to insider trading.” That sort of thing. As a matter of science fiction it seems pretty trivial for an omniscient superintelligence to find cool ways make money. “Charge retail customers $20 per month to access the superintelligence,” what, no, obviously that’s not the answer. On a pure science-fiction suspension-of-disbelief basis, this business plan is perfect and should not need any updating until they finish building the superintelligent AI. Paying one billion dollars for a 0.2% stake in whatever God comes up with is a good trade. But in the six years since announcing this perfect business plan, Sam Altman has learned [2] that it will cost at least a few trillion dollars to build the super-AI, and it turns out that the supply of science-fiction-suspension-of-disbelief capital is really quite large but not trillions of dollars. Trillions of dollars is going to tap out every source of venture capital, every investor who is willing to put in money now for an uncertain but potentially huge future return. To get trillions of dollars you will need boring debt investors, people who need their money back, people who are willing to put in money now for 9% a year and repayment in five years. Very different psychology. Those people don’t want 0.2% of God; they want a financial model, with numbers, and a reasonably likely path to getting paid back. Intriguingly OpenAI’s AI models these days are already pretty good at producing compelling, credibly formatted, deeply researched documents, so, you know, the vision has kind of been fulfilled? Like could OpenAI’s finance staff open up OpenAI’s top-end current models and type in “tell us how to make money” and get a good answer? Maybe? Anyway they’re generating an answer somehow. The Financial Times reports: OpenAI is working on new revenue lines, debt partnerships and further fundraising as part of a five-year plan to make good on the more than $1tn in spending it has pledged to create world-leading artificial intelligence. OpenAI is planning on deals to serve governments and businesses with more bespoke products, creating more income from new shopping tools, and new sales from its video creation service Sora and AI agents, said multiple people familiar with the start-up’s efforts. ... It is also weighing ways to cash in on its intellectual property by developing new AI infrastructure, making forays into online advertising and plans to launch consumer hardware products, including a new AI-powered personal assistant device, with former Apple star designer Jony Ive. These ambitious plans will need to become reality if OpenAI is to meet its liabilities, as the group has made funding commitments that dwarf its income. … One senior OpenAI executive said “[investors] expect you to have a five-year model”, but added “right now I’d say there’s lots of fuzz on the horizon, and as it gets closer and it’s going to start to take real shape”. … Greg Brockman, the company’s president, last week said recent spending commitments would pay for themselves: “If we had 10 [times] more compute [computing power], I don’t know if we’d have 10 [times] more revenue, but I don’t think we would be that far.” Right with 10 times the compute maybe you get God and money just rains from the sky. Meanwhile, though, “new shopping tools” and “forays into online advertising.” In other business news: ChatGPT will allow a wider range of content — eventually including "erotica" — now that OpenAI has completed work to enable the chatbot to better handle mental health issues, CEO Sam Altman said Tuesday. … "As part of our 'treat adult users like adults' principle, we will allow even more, like erotica for verified adults," he added, though that would come around December. I began this section with a jokey maximalist vision of AI, “create God,” “an omniscient superintelligence,” that sort of thing. The jokey minimalist vision of AI is probably “ChatGPT is a blurry JPEG of the web”: Modern AI systems are approximately a synthesis of all human knowledge and communication, but given the way computers work, that means especially a synthesis of the internet, which is where you get the bulk of machine-readable human knowledge and communication. Ryan Broderick writes: “Think of ChatGPT as a big shuffle button of almost everything we’ve ever put online.” I once wrote about asking ChatGPT to pick stocks: If you ask a modern publicly available large language model which stocks to buy, it will in some sense draw on all of human knowledge and its own powerful reasoning capacity to tell you which stocks to buy. But, among all of human knowledge, it might give extra weight to the knowledge on Reddit. And the knowledge on Reddit about what stocks to buy is “meme stocks.”
You can apply similar reasoning here. In a science fiction story, if you invented a superintelligent robot and asked it how to make money, it might come up with cool never-before-seen ideas, or at least massive fun market manipulation. But in real life, if you train a large language model on the internet and ask it how to make money, it will say “advertising, affiliate shopping links and porn.” That’s the lesson the internet teaches! Anyway I do not actually think that ChatGPT is coming up with OpenAI’s business plans. ChatGPT is trained on the internet, but of course these days so is everyone else. [3] “We have a tech company, let’s do ads and shopping and porn” is not a thought process that is limited to large language models. The 24-hour stock exchange, which will offer stock trading 23 hours a day, now offers stock trading 16 hours a day: In a historic milestone for the financial industry, 24 Exchange today announced that trading has officially commenced on 24X National Exchange (the "Exchange"), the first national securities exchange approved by the U.S. Securities and Exchange Commission (SEC) to offer 23-hour weekday trading of U.S. equities under full regulatory oversight. 24X National Exchange offers live trading of U.S. equities from 4:00 a.m. to 8:00 p.m. ET on weekdays, providing unprecedented access to U.S. equity markets for institutional investors and retail investors worldwide via broker-dealers that are members of this SEC-licensed Exchange. 24X National Exchange addresses growing demand for extended-hours trading among global investors, particularly in Asian and other regions where time-zones make traditional U.S. market hours difficult to access. With its early-morning (U.S. pre-market) and evening trading hours, the Exchange provides increased flexibility for trading strategies, improved liquidity, and more opportunities to react to global economic, corporate, or policy news outside of standard U.S. exchange operating hours. They’re getting there. I will say that one theme around here is that … everything is getting dumber? I don’t know that that’s actually much of a theme, sorry. But I wrote in April: Another model of capital markets is that people like gambling, so introducing some extra volatility makes markets more fun and exciting and gives people what they want. How much should you save for retirement? Should you borrow money to build a new factory? Boring! Boring! Don’t ask those questions! Ask more fun questions like “should I YOLO all my money on GameStop call options?” On this model, economic policymakers should lurch drastically from one policy to another, because that will make things more exciting for their audience. It will also get the policymakers more attention, and attention is the most valuable thing in the world.
And we have spent a lot of time this year talking about how financial markets are being swallowed by sports gambling. And meme stocks are back, and crypto is mainstream, and so forth. A decade ago, it was possible to think that investing would gradually become duller and more efficient and more automated, that people would just let robots and professionals manage their retirement savings and would not look for excitement in their investment portfolios. Now that sounds crazy: Of course people want to invest in single-game sports parlays and zero-day options and triple-levered exchange-traded funds and meme coins. And the market gives them what they want. What they want is excitement, volatility, binary outcomes, jump discontinuities, just general chaos, because that is more entertaining than slow and steady compounding. I don’t think that 24/7 or 23/5 or 16/5 stock trading is exactly part of this trend: Late-night trading does give US retail investors more time to trade while drunk, but as 24 Exchange says, it is probably mostly about Asian investor demand. (Also most of the big exchanges already have extended sessions and have for a while.) Still I think there is a connection. We have talked a few times about glitches in after-hours trading. For instance, last year, Lyft Inc. had a typo in an earnings release, saying that it expected margins to expand by 500 basis points when the actual number was 50. The incorrect press release went out at 4:06 p.m.; it was corrected at 5:51 p.m. In the intervening time, the stock jumped as much as 67% on fairly heavy volume; it gave up much of those gains when the correction came out. Or last week, a meme-stock ETF launched; after hours, it traded at prices that were wildly disconnected from its net asset value. My crude explanation was: “Retail investors got excited to buy the MEME ETF, so they did. During regular market hours, they bought the MEME ETF from market makers, and arbitrage relationships kept its price in line with its net asset value. At 4 o’clock, the market makers went home and the price of the MEME ETF went crazy: Lots of retail investors wanted to buy it, they didn’t care about the price, and nobody was bothering to arbitrage it to keep the price in line with net asset value.” The two points here are: - Traditionally the US stock market was closed 17.5 hours per day so that most material news could be released outside of market hours. If you have earnings, you announce them at 4:05 p.m. If you do a merger, you finalize it over the weekend and announce it at 6 a.m. on Monday. The market has time to digest the new information; investors decide what it means and update their models and decide whether they want to buy or sell stock. Then they all show up at the market open and conduct an orderly auction to set the new opening price. What this means is that, as a matter of market time, everyone gets the information at the same time: Even if you do not have high-speed computers scraping earnings releases and using artificial intelligence to analyze them for trading signals, you can still get the press release when you think of it, read it at your leisure, and decide what to do about it. Getting the press release five seconds or five minutes later than other people doesn’t matter, because you have 17 hours before you can do anything about it. By the time the market opens, everyone has the same basic information.
- During the 6.5 hours a day that the market was traditionally open, all the professional investors are there. All the arbitrageurs are doing the arbitrages; all the computers are turned on. Any new information that does come out during market hours is swiftly and smoothly incorporated into prices. Professional investors compete to make markets more efficient. One thing this means is that if you decide to buy 10 shares of an S&P 500 index ETF at 11:14 a.m., you will pay a price for that ETF that (1) pretty much exactly reflects the prices of the component stocks of the index and (2) also reflects the market’s consensus of the value of those component stocks. The market might be wrong, but you will be investing in the stock market at a price that is the aggregated best estimate of sophisticated investors for what the “correct” price is. If you are saving for retirement, buying the stock market at the correct price — and getting the expected return from the growth of corporate profits — is the boring normal thing you want.
But if you trade stocks at 4:30 a.m., or 7:30 p.m., you are getting a different proposition. News comes out — perhaps an earnings release at 4:05, or perhaps an Asian earthquake at 3 a.m. — and you get to decide how to react to it. Other people — retail investors like you, investors in Asia, hedge funds who haven’t gone home — are also deciding how to react, in a messy, sequential way. The stock will bounce around: Some smart investors will think the news is good and push up the price, other smart investors will think it’s bad and push down the price, and then dumb investors will pile in and push the price around some more. Meanwhile fewer smart professional arbitrageurs and market makers might be logged in, so the prices will move more rapidly and more randomly than they will during the day. There is more scope for skill, and much more scope for luck. You will probably get to buy stock at the wrong price! Too high or too low! Exciting. If you are saving for retirement by buying index ETFs, you’ll probably still do it during the day and get normal prices and it’s fine. But if you want some chaos, you can probably get more of it late at night. And people seem to want chaos. One purpose of financial markets is to incentivize people to find out information and incorporate it into prices. If you buy an iPhone and think “this phone is cool, I should buy Apple stock” and you do and the stock price goes up, you are doing your small part to allocate capital to its best uses and make stock prices more informative. There are more sophisticated versions of this. If you work at a hedge fund, you might analyze credit card data or photos of parking lots to figure out whether a company’s sales are going up this quarter. If enough people are doing that, then information from the world — about who is driving to what stores, etc. — is nearly instantaneously incorporated into asset prices, making markets more efficient. Then there is the bad version, which is insider trading: You are an executive of a company, you know that earnings will be bad, so you sell your stock. Or you are a hedge fund analyst, you bribe an executive of a company to tell you that earnings will be bad, so you short the stock. This also incorporates information into asset prices and makes markets more efficient, but people don’t like it. It’s unfair, it discourages people who don’t have inside information from trading, and it creates bad incentives for insiders. It is generally illegal, in the stock market anyway. And then there is the just-on-the-right-side-of-the-line version, which is web scraping. Companies often put out material news on their websites. If you work at a hedge fund, you might set up some automated system to immediately download, analyze and trade on whatever news the company puts on its website; or you might just sit in front of your computer and constantly refresh the website looking for news. (Particularly when the company is scheduled to announce earnings.) Every so often, the news will show up on some section of the company’s website before it is officially announced: There will be some page like “xyzcorp.com/investor/earnings_2025q2.html,” and the earnings will show up there before they are on the homepage or the news wire. If you know the URL format of the company’s earnings releases, you might type “xyzcorp.com/investor/earnings_2025q2.html” into your browser a few minutes before the earnings are scheduled to be released, and you might occasionally get lucky. This is a real thing that has occasionally worked for corporate and government news releases. It is not, in isolation, all that good for the world: Capital allocation is not made much more efficient by some hedge fund analyst refreshing a website a minute before everyone else. But you probably should not think of it in isolation. The point here is not that markets are made more efficient by hedge fund analysts refreshing websites; the point here is that markets are made more efficient by hedge fund analysts working 80-hour weeks to find every possible legal edge to understand companies more deeply and quickly than everybody else. Some of the edges they find will involve deep fundamental insight into corporate finance or weather or whatever. Others will involve quickly and automatically incorporating price signals from one market into the pricing of different assets: A high-speed exchange-traded-fund arbitrageur is not doing much fundamental analysis, but she is making markets more efficient by rapidly propagating other people’s fundamental analysis. And some of the edges will be dumb low-hanging fruit like refreshing the website. But you can’t have a rule like “you can only use good fundamental edges.” You let people compete to find whatever edges they can, because in the aggregate that competition makes markets more efficient. Anyway. I wrote yesterday about rumors of insider trading on Polymarket about who would win the Nobel Peace Prize. My basic point was that there is only so much Nobel Peace Prize insider trading that you can do (the alleged insider trader here made about $50,000), because who will trade against you? I said: Are millions of recreational gamblers really going to want to take a flutter on which democracy activist will win the Nobel Peace Prize? Is some quantitative hedge fund going to build a statistical model, using only public information, to predict the Nobel Peace Prize winner, and wager tens of millions of dollars on that model? Really the only person who had a good reason to bet on this market was this supposed insider trader. What you want, as an insider trader, is a lot of uninformed traders who nonetheless have a good reason to bet against you. Football prediction markets have that. Nobel Prize prediction markets, so far, don’t.
But here is an account on X of an interview with the alleged insider trader, which suggests that I was wrong about this in the most pleasing possible way. This trader apparently didn’t have inside information, in the traditional bad sense of like bribing a Nobel committee staffer. Instead, web scraping: The Nobel site runs on WordPress. Like many WordPress setups, it has an XML sitemap that lists every indexable page, even ones not yet public. If someone were monitoring this sitemap, they could easily notice a new page appear, something like "http://nobelprize.org/prizes/peace/2025/machado/facts/" Most likely, someone wrote a bot or script that watched the uploads folder or sitemap for new files. When this image appeared, they connected the dots. On the Nobel website, the official portrait of Machado at https://nobelprize.org/uploads/2025/10/1-2_3-68e8b33bd36fa-scaled.jpg carries a “last-modified” timestamp in the file’s metadata of 07:18 GMT on October 10, over 1.5 hours before the announcement. This picture was loaded into the official announcement page of Nobel “http://nobelprize.org/prizes/peace/2025”, a link you could guess by switching in 2025 for 2024 Etc. The point here is not so much “perhaps this trader did not have inside information.” The point is: Perhaps someone was incentivized to build, if not quite “a statistical model using only public information,” at least, like, an automated feed of public alternative data on the Nobel Peace Prize winner. The point is that doing informed but non-insider trading on the Nobel Peace Prize winner is (1) possible and (2) possibly lucrative enough that people will do it. In a dumb, degenerate, not-especially-fundamental way, but still. The point is that the Nobel Peace Prize prediction market is working the way it is supposed to: It incentivizes people to find out information and incorporate it into prices. Morgan Stanley Stock Traders Beat Goldman in Record Quarter. BofA Bankers Beat Estimates as Dealmaking Boosts Earnings. Wall Street Is Firing on All Cylinders, Fueled by Deals and Trading. Bank of England loosens bonus rules. BlackRock’s GIP Buys Aligned Data Centers in $40 Billion Bet. A Giant New AI Data Center Is Coming to the Epicenter of America’s Fracking Boom. Blackstone Joins Race to Bring Private Assets to 401(k) Market. S&P Global agrees $1.8bn deal for private markets data provider. Trump Says Argentina Bailout Depends on Milei’s Party Winning Elections. EU Antitrust Enforcer Fines Gucci, Chloe, Loewe More Than $180 Million After Pricing Probe. Chinese Criminals Made More Than $1 Billion From Those Annoying Texts. The overnight anomaly in levered Tesla ETFs. “Travis Kelce’s new restaurant, 1587 Prime — where, the website reads, ‘legacy isn’t just honored, it’s transformed into something entirely new,’ which translates to ‘it’s a steakhouse.’” 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! |
