Monopoly Round-Up: How New Dealers Would Have Popped the AI Bubble

Lots of monopoly news, as usual. There was drama in the Paramount-Warner merger, record and unexpectedly high corporate profits are contrasting with sour consumers, and the most important Senate primary around monopoly questions takes place on Tuesday.

But before getting to the full round-up, I want to start by discussing some market-rigging that took place around AI this week, and what it means. The tldr here is pretty simple – it’s time to force big AI firms to open their finances to the public. And I kind of hit upon this idea from two different directions.

Let’s start with what happened in the markets. A $45 billion AI-focused hedge fund called “Situational Awareness” blew up, losing the vast majority of its valuation before being forced to sell its positions to Citadel. The founder of Situational Awareness is a 25 year-old former OpenAI employee named Leopold Aschenbrenner. Aschenbrenner made the classic mistake of betting with with borrowed money on volatile stocks (including a big chunk of Anthropic, which is privately traded.). After his public stocks got hit, Aschenbrenner got calls from the banks to pay back money; he had to sell to raise the cash, and the fund collapsed.

Aschenbrenner’s thesis was simple. He believed that he is among a select few who understand that AI is going to become unimaginably powerful, and he invested accordingly. For a time, he was the market’s hot hand, mimicked by an entire generation of young speculators.

But more important is that he is part of a network in Silicon Valley that is manipulating AI stocks with cult-like claims about social transformations larger than the industrial revolution. Aschenbrenner’s an effective altruist, the cult suffusing Silicon Valley and certain parts of the D.C-based Abundance world. He even got his start at Sam Bankman-Fried’s FTX. He’s married to Anthropic CEO Dario Amodei’s chief of staff, and his roommates are AI booster podcaster Dwarkesh Patel and Anthropic’s Sholto Douglas. To give you a sense of the insularity and weirdness of this world, his wedding included a “colloquium to discuss ideas in panels and breakout sessions.”

Now, AI hype is routine, such that we have all gotten really really tired of it. OpenAI CEO Sam Altman is well-known for insane claims; he recently said that “We are close to creating the genie that can grant any wish.” A few days ago, OpenAI announced its new model solved ten open math problems that had bedeviled the mathematics community, with hype suggesting that it’s Fields medal-level impressive. As with most such announcements, the puffery has outrun the truth, with at least one of the claims already disproven.

Altman is famous, a global icon even, but it’s the Sholto/Patel/Aschenbrenner axis that do the routine work of making the hype seem plausible in financial terms. In April, Patel co-authored an essay on how the profit margins for Anthropic’s sale of computing power to run AI models, known as “inference,” increased “from 38% to over 70%” in just a few months. That’s not a big dumb claim like “we are creating a genie,” but it’s a narrower yet equally significant claim signaling to financiers to keep investing in a business model becoming insanely profitable.

These guys tell a lot of stories about AI, including financial ones. Is this margin story true? Maybe. But we don’t know. And we should, as this information matters. Trillions of dollars of investment, in fact a non-trivial chunk of the U.S. economic future, rests on this financial data. But unlike a more abstract and unknowable question, like “what will we use AI to do in a few years,” this one is answerable. Insiders already know how much Anthropic is selling to corporate America, and how much it costs to provide it. It’s just not public information.

And so we are forced to rely on skeptics without access to this inside information, such as Ed Zitron. Zitron is probably the most prominent skeptic of the claims of AI firms and of their financing, and he gets a lot of criticism from the industry. Here he is arguing that much of what is happening is a hidden financial shell game. Is he right? Well, insiders at Anthropic, Nvidia, Google, and OpenAI know. They could disprove Zitron easily, if the data suggested he was wrong. But they don’t. And this game is high-stakes.

This financial opacity is also distorting our politics. A few days ago, a Congressional candidate called me and asked if I had “antitrust” ideas for AI. He doesn’t love the Bernie Sanders approach, which is to have the state take 50% ownership of big AI firms and put them in a sovereign wealth fund, but he agrees with the basic premise that the government should regulate this technology to be good for people not bad for them. Most of us do.

As an anti-monopolist, I’ve been studying this space for years, so it’s weird that I don’t have an actual policy solution here. Generative AI is a result of a specific type of industrial policy. The state sanctioned big tech firms to gain monopoly profits, fostered uniquely permissive copyright and privacy rules for those firms, and prevented them from having to accept liability for harming their customers. With the resulting profits, these companies built data centers, usually with public subsidies, that suck up electricity and water and are generally nuisances.

And yet, it’s not totally clear how to govern this set of technologies. The market structure is confusing, and it also keeps changing. Chinese open source models seem to be disciplining the pricing power of the top firms, Nvidia is funding a lot of its customers, and the big tech corporations are also investing tens of billions of dollars into OpenAI and Anthropic, which then send that money back to big tech firms in the form of cloud computing contracts. There is even a pretty good argument that it is impossible to sell compute at scale, because if a lot of big requests come in at once, your costs could jump in unanticipated ways. So discussions of market power are hard to form in a coherent way, simply because we don’t have the financial information to know what these business models really are.

Eventually, I asked myself, why am I guessing about all of these things that are obviously a matter of public importance? And so I started thinking about how we really don’t know much about the financial inner workings of the big AI companies. They know everything about us, since they have sucked up all our data. But their finances are a black box, either because OpenAI and Anthropic aren’t public companies OR because the Google, Amazon, Microsoft, and Nvidia are so big and complex that they can simply hide their AI revenue through cross-subsidization. There are also a good number of shell companies and off-balance sheet instruments hiding what is going on. We know they are raising and spending economy-shaking amounts of capital, but beyond that, it’s guesswork.

So I thought about this candidate’s question, and the answer I’ve come up with is to mandate these companies to just open their books. And I think that in and of itself would have important consequences. In fact, it’s kind of weird OpenAI and Anthropic are not public companies. Financial openness by big firms was the bedrock of how New Dealers in the 1930s dealt with corporate America, because the opaque financial tactics of financiers in the 1920s had led to a bubble and a crash. In 1933, Congress passed the Securities Act, which forced public companies to disclose material financial information publicly.

For eighty years, if a company passed a certain threshold – most recently 500 shareholders and $10 million in assets – it had to register with the Securities and Exchange Commission (SEC) and disclose. Through the explosion of high-tech firms, from Intel to Microsoft to Google, one important step was to have a firm go public, simply because it had to as its stock-owning employees multiplied. This dynamic fostered openness, as soon as a company grew to a certain size, it usually ended up having to disclose a lot more data by dint of the registration threshold.

In 2012, however, the Obama administration worked with the GOP and a set of Democrats to roll this system back through a law called the JOBS Act. Obama argued the New Deal SEC rules were outdated, that “laws that are nearly eight decades old make it impossible for others to invest” in startups. The JOBS Act raised the number of allowable outside investors in private companies to 2000, and excluded employees with stock from that number.

The result was an explosion in private markets, and firms like WeWork and SpaceX. I wrote about this dynamic in 2019, walking readers through how less regulated private markets enabled the founder of WeWork to keep manipulating the valuations of his company upward. WeWork blew up when it tried to go public, and released its S-1 form with all that batshit stuff about Adam Neumann. And SpaceX has fallen by 50% since it went public because everyone can see the financial information clearly. Transparency works.

Of course just repealing the JOBS Act doesn’t get the whole thing done because you still have giant companies like Google, Amazon, et al that are not disclosing what seems like material information about their lines of business. So you’d also need better regulation. A good SEC could do better enforcing its disclosure requirements around material lines of business and material revenues and risks. The Public Company Accounting Oversight Board could tighten inspections and enforcement around auditors of the big hyperscalers, and stop them from doing all the off-balance sheet nonsense.

Fundamentally, there’s something powerful about saying “AI firms need to open their financials to the public.” Without information on what they are selling and whether it’s profitable, we just don’t know enough about the markets to govern the technology or tell whether we’re in an economy-shaking bubble. We’re in a weird and confused netherworld, of guesses and spin, organized by gamblers.

That’s no way to run an economy.

And now, the rest of the monopoly round-up. There’s a glorious hate read about the awful big law firm Paul Weiss, and how the worst people in the world got their comeuppance. Plus, on Tuesday, Democratic anti-monopolist Abdul El-Sayed will face establishment Dem Haley Stevens in a Michigan Senate primary that could set the tone for the 2028 Presidential race. There’s also important news on AI and price-fixing, Paramount-Warner, and the Fed quietly proposing to allow banks to more easily lend to their own executives. What could go wrong?

All after the flip.

Read more