Prediction Markets Are the New Public Markets

In the 1930s, the problem with US stock markets was that investors didn’t get enough information. The solution was new rules that created the US Securities and Exchange Commission and required companies to disclose various information — audited financial statements, updates on material events, etc. — if they wanted to sell stock to the general public. Companies that didn’t want to raise money from the general public didn’t really have to disclose anything, but the general public had most of the money, so most companies that wanted to get big had to go public and disclose the required information.

In the 2020s, the problem with US stock markets is apparently that not enough companies want to do that disclosure. “Private markets are the new public markets,” I often say: It is relatively easy for hot tech startups to raise lots of money from institutional investors, without ever selling to the general public, and so many of them do. They don’t go public, they don’t file disclosures, and they just raise money in private markets.

This arguably creates two problems:

1. Hot tech startups create a lot of value, and if they don’t sell shares to the general public, then the general public doesn’t get any of that value. SpaceX and OpenAI and Anthropic all have gigantic valuations; all of their growth so far has been captured by employees and founders and venture capitalists and other big investors, not by the general public. [1]
2. There are good social externalities in having companies be public and disclose their information. Society — journalists and politicians and activists and ordinary people — can learn important stuff by reading public-company disclosures; those disclosures are not of interest only to shareholders.

A lot of people worry about the first problem (not so much the second) and try to think of solutions. One set of solutions involves making it harder for companies to stay private (so they have to go public earlier and sell stock to the general public); another set of solutions involves making it easier for companies to go public. “Making it easier for companies to go public” often means reducing the standard disclosure requirements: It’s easier for a company to be public if it doesn’t have to disclose as much about its finances. We talked yesterday about an SEC proposal to make quarterly financial reporting optional: If companies could report results only every six months, that would be easier, and more of them might go public. Today the SEC proposed some more changes to reporting rules to make being public easier and “Make IPOs Great Again” 

But there is another, more radical, solution, which is to get rid of the whole disclosure regime and let the general public buy shares of private companies without any disclosure. Companies wouldn’t have to “go public”; they wouldn’t have to file financial statements or disclose material events; they could just remain private. But ordinary investors could buy their shares anyway, without the disclosure requirements that have become standard over the last century.

Nobody ever quite says this, because it sounds bad. They just try to do it, by creating new ways for ordinary investors to trade private company shares freely without any disclosure. Generally the way to do it involves letting ordinary investors trade some instrument that is economically equivalent to private company stock, or at least pretty close, rather than actually trading the private stock.

In recent years the standard name for this approach has been “tokenization.” There was a big boom in crypto, and a lot of people had a vague sense that calling something a “token” made it immune from securities laws, so people went around saying “if we issue tokens on private companies, that will let people buy exposure to their stock without requiring any securities-law disclosure.” I wrote about this theory last year. It has some quite prominent proponents — Vlad Tenev of Robinhood, Larry Fink of BlackRock — who quite explicitly argue that tokenizing private-company stocks would allow ordinary investors to buy shares of private companies. “Democratizing” finance, this is called. But I wrote:

This solution doesn’t work, yet, in the US. You can’t just sell “tokens” of private-company stocks (or private credit loans, or private equity funds, etc.) to the general public, in the US, without disclosure, yet. But a lot of big players in finance are advocating for it, the regulatory environment seems pretty receptive, and you can understand why. The general public wants to buy private investments, intermediaries want to sell them, and the disclosure rules stand in the way. Saying “we should get rid of the disclosure rules” sounds bad, retrograde, greedy. Saying “tokenization” sounds good, modern, cool.

That was last year, when “tokenization” sounded good and modern and cool. (Though “the Securities and Exchange Commission is expected to release its so-called innovation exemption for tokenized stocks as soon as this week.”) Now it is 2026, and “prediction markets” is the new cool modern term. Here’s a press release from Polymarket:

Polymarket, the world's largest prediction market, today announced the launch of the first prediction markets tied to private company performance and milestones. These new offerings give individuals exposure to some of the most sought-after private companies for the first time, while providing a new real-time signal for institutional investors on how private markets are unfolding. Through an exclusive agreement, Nasdaq Private Market, a leading provider of liquidity and investment infrastructure for the private market, will serve as the resolution data provider for private company markets on Polymarket.

More value is being created in private companies today than at any point in modern history. Nearly 1,600 unicorns globally now hold more than $5 trillion in cumulative value, yet access has been largely reserved for institutions and high-net-worth investors – leaving the vast majority of individuals on the sidelines of the value creation that happens before a company goes public.

"Prediction markets are one of the most powerful tools we have for democratizing access to financial information and opportunity," said Shayne Coplan, Founder and CEO of Polymarket. "Today’s launch brings that power to one of the last frontiers of financial markets that retail participants have never been able to access. For the first time, anyone can engage with the outcomes driving value at the world's most consequential private companies."

Individual investors can’t buy shares of OpenAI or Anthropic or SpaceX, because those companies are not public. But now they can bet on “Will OpenAI’s valuation hit __ by December 31,” for various numbers that might go in that blank, or similar markets for Anthropic and SpaceX and other private companies. (These binary bets are not quite as good as actually buying stock, but give them time.) “Democratizing access to financial information,” here, means betting on stocks of companies without public disclosure. You get to trade more stocks, but you get less information about them.

Again, technically, this solution doesn’t work in the US, because technically US investors are not allowed to use Polymarket’s main site. Nobody has ever seemed to care much about this, and Polymarket did launch its US app last week. These markets are probably “security-based swaps,” which should in theory make it difficult for a regulated US prediction market to list them, but again nobody cares very much. “Anyone can engage” with these markets, says Coplan. The fact that Nasdaq — a big actual US stock exchange — is teaming up with Polymarket suggests that this is all more or less fine by regulators. “We should make it easy for ordinary retail investors to bet on private company stocks” seems, at this point, pretty uncontroversial. 

We seem to be reaching the end of securities regulation. For almost 100 years, the bargain in US securities law was that companies could sell stock to ordinary individual investors only if they publicly disclosed material information. The public could trade public stocks, which made public disclosures, but not private stocks, which didn’t. Everyone seems tired of that bargain and now it is breaking down.

Here are some shareholders of Anthropic PBC: Microsoft Corp., Nvidia Corp., Amazon.com Inc. and Google (part of Alphabet Inc.). Anthropic is a private company, and you cannot buy its shares on the stock exchange, though now I guess you can bet on them on Polymarket. Microsoft, Nvidia, Amazon and Alphabet, however, are public companies, and you can buy their shares on the stock exchange. You can get some indirect exposure to Anthropic: You can buy shares of Microsoft or Nvidia or Amazon or Google, and then you will own shares in a company that owns shares in Anthropic.

One problem with this is that all of those companies also own other stuff, besides their Anthropic shares. Some of that stuff is unrelated to Anthropic; some of it is negatively correlated with Anthropic. (Google and Anthropic are rivals in building AI models, etc.) Buying Nvidia stock is not a clean bet on the future of Anthropic; it is muddied by Nvidia’s actual business. 

It would be tidier to buy shares of a company that only owned Anthropic shares. Then, if you wanted some exposure to Anthropic, you could just buy shares of that company. 

Of course Anthropic might not like that. Anthropic’s stock is subject to transfer restrictions: If you own Anthropic stock, you can’t generally sell it without Anthropic’s permission. We talked last week about Anthropic’s attempt to enforce those restrictions more strictly. One traditional way around transfer restrictions is the forward contract, where instead of selling you Anthropic stock today, I promise to deliver it to you when Anthropic goes public (and the transfer restrictions are removed). Private companies’ transfer restrictions generally prohibit forward contracts, but people don’t always take that seriously: After all, once the company goes public and the transfer restrictions are removed, what does the company care about people who sold forwards years ago? How will it even find out? But Anthropic updated its website to say, in effect, “no, if you buy a forward, you might not get the stock,” which got a lot of people worried.

Yesterday Bloomberg’s Yazhou Sun and Katia Porzecanski reported on the other threat in Anthropic’s restrictions:

Anthropic PBC issued a stern warning on its website last week about unauthorized sales, taking the unusual step of naming eight firms whose offerings would be considered void. It also expressly prohibited investors from buying shares through special purpose vehicles, a common tool to raise financing. …

As secondary sales grew, so did SPVs, now a standard part of financing for many pre-IPO companies and a way for smaller investors to buy in. An early Anthropic backer, Menlo Ventures, pooled its own $500 million check into an earlier funding round using an SPV, according to an SEC filing first reported by Business Insider.

But as funding rounds have grown, SPVs have become complicated and murky, with little end clarity for the buyer and reduced transparency for the startup.

“Anthropic does and should have the right to control its cap table and shareholder base,” said Matt Murphy, the Menlo partner who led the firm’s investment. He said his firm didn’t use the kind of SPVs banned by Anthropic.

“Unauthorized SPVs are not in the company's best interest and can often be downright shady,” he said. “Buyer beware.”

An SPV is the tidy solution I described above: It’s a company that owns shares of Anthropic. You can’t necessarily buy shares of Anthropic, but you can buy shares of the SPV.

Again, Anthropic doesn’t like it, and says that its transfer restrictions prohibit it. “Anthropic does and should have the right to control its cap table and shareholder base.” But, sort of. Anthropic doesn’t get to control Alphabet’s shareholder base. Anthropic can put transfer restrictions on its stock; it can say “you can’t sell this Anthropic stock to anyone else.” But it is harder for Anthropic to put transfer restrictions on its shareholders’ stock: If a company invests in Anthropic, Anthropic can’t tell that company who can own its stock. (Anthropic can try to prevent SPVs from buying its stock; it can limit its investors only to people who promise not to do SPVs. But there might be gray areas.) You can see why people would think that SPVs could be a workaround to transfer restrictions.

Incidentally. If you are an investor who owns $90 million of Anthropic stock, and you want to sell now while the selling is good, what should you do? The traditional approaches are all transfer-restricted:

• You could just sell your shares to someone else, but you need Anthropic’s permission.
• You could sell your shares under a forward contract, but Anthropic says that’s not allowed. 
• You could package your shares into an SPV and sell shares of the SPV, but Anthropic says that’s not allowed either.

But now there’s a new approach. You can sell 100 million shares of “Will Anthropic’s valuation hit $1 trillion by December 31?” on Polymarket for about $91 million. If the valuation hits $1 trillion by Dec. 31, you have to pay out $100 million — but your shares will be worth at least $100 million in that scenario, so you’ll be hedged. [2]  If it doesn’t, you keep the $91 million (and the shares!), so it’s even better than selling your stock now. Obviously this doesn’t work yet, because that market is brand new and the volume is measured in the low thousands of dollars. But give it time.

An important job of a banker is to offer clients loans. If an investment banker or commercial banker from Citigroup Inc. shows up at your office to chat about merger ideas or your new factory, and the chat goes well, you will probably ask her: “Well, how are we going to pay for this?” And you will expect her to have an answer. At some point back in the sands of time perhaps the answer was “we will lend you the money,” but in modern finance the more standard answer is “we will be happy to arrange a package of loans for you” or “we will be happy to arrange a bond offering for you.” The bank will facilitate the deal, but the bank — even a giant bank like Citi — won’t lend you all of the money itself. It doesn’t have that kind of money.

These days, an important competitor to those two types of financing — syndicated loans, bonds — is private credit. And so, to provide good customer service, the banker really has to answer your question by saying “we will be happy to arrange a package of loans, or a bond offering, or a private credit direct loan.” (I have written before that the banks will have to add another column to their pitchbook pages.) But whereas arranging a broadly syndicated loan or a bond offering takes time, the advantage of private credit is that it can be done quickly: The private credit firm already has the money and can just give it to you. There is not a long marketing and syndication process. For really good customer service, the banker has to say, like, “we can give you a private credit loan, here are firm terms, we can sign right now.”

But the bank doesn’t have that kind of money! Private credit funds have that kind of money. And so what you want is for the bank to be sort of a conduit for a private credit fund, a customer-service front-end for the private credit fund. The bank has bankers and clients and a marketing apparatus; the private credit fund has fewer employees and less customer surface area, but it has the money. There is something obviously weird about this — don’t banks have money? — but it’s the world we live in. One approach is for the bank to raise its own private credit funds — many big investment banks would like to be alternative asset managers themselves — but the other approach is to team up. We talked in 2024 about Citi’s arrangement with Apollo Global Management, in which Citi pitches some loans and Apollo makes them. Because Apollo has the money. And Citi doesn’t. You know.

Yesterday Bloomberg’s Claire Ruckin and Silas Brown reported:

Citigroup Inc. and BlackRock Inc.’s private credit unit HPS Investment Partners have struck an agreement to collaborate on direct-lending deals across Europe, targeting as much as €15 billion ($17.5 billion) of financings over the next five years.

The initiative — Citi/HPS Private Capital Program — will focus on sub-investment grade debt for corporate and private equity clients in continental Europe and the UK, according to a statement seen by Bloomberg News. The Wall Street firms will work together on both senior and junior credit options and eventually expand the collaboration to deals in the Middle East.

“It’s HPS’ capital, credit decision and ultimately their loan, but it’s an origination and structuring that we bring to them, and then jointly figure out whether there’s a deal to be done,” Citigroup’s co-head of global debt capital markets John McAuley said in an interview.

The tie-up builds on a strategy that Citigroup first unveiled in the US two years ago, and highlights the growing convergence between traditional banks and private credit managers. Long considered competitors in corporate and sponsor-backed lending, the two are now forging ties as banks try to preserve fee income without using additional balance sheet capacity in the face of higher capital requirements.

It remains so weird to me. HPS has $193 billion of assets; Citi has $2.8 trillion. Citi has higher capital requirements than it used to, but its ratio of equity to assets is something like 7%, while most private credit funds have something like 50% equity to assets. Private credit funds have much less money and much higher capital ratios than banks do, but they’re the ones making the loans.

OpenAI was founded in 2015 as a nonprofit organization with a mission “to ensure that artificial general intelligence benefits all of humanity.” It remained a nonprofit until 2019. In 2025, it became a for-profit company, although the old nonprofit retains a significant stake in the company.

What happened between 2019 and 2025? Ehh. For those six years, OpenAI was still technically a nonprofit, but it had a for-profit subsidiary that raised money from investors and did all of the business. This was a weird structure: The subsidiary was actually a “capped-profit” company, it went around telling investors that “it would be wise to view any investment in OpenAI Global, LLC in the spirit of a donation,” and for like 48 hours the nonprofit’s board fired Sam Altman, OpenAI’s co-founder and CEO. “OpenAI Is a Strange Nonprofit,” I wrote in 2023, one with an $86 billion valuation. Was it? During those six years, was OpenAI mostly a nonprofit, or mostly a for-profit company?

This strikes me as a boring question — it was a little of both — but it turned out to matter a lot. In 2024, Elon Musk — who co-founded the OpenAI nonprofit but left in around 2018 — sued OpenAI, Altman and other co-founders, arguing that, as he put it, they “stole a charity”: OpenAI was a nonprofit, Altman converted it into a for-profit, and now its employees have billions of dollars of stock in a valuable private startup, a startup that they built on charitable donations from Musk and others. 

And the critical question in that lawsuit was: When did they steal the charity? Musk’s legal theories have two- and three-year statutes of limitations: If you want to sue for fraud, breach of contract, etc., you have to do so within two or three years after you become aware of the fraud, breach, etc. OpenAI argued that it became more or less a for-profit company in 2019, so whatever problems Musk had with the transition, he should have raised them then. From its court filings:

Musk was aware of the conduct purportedly constituting these alleged breaches no later than 2019, well outside the limitations period. OpenAI, Inc. publicly announced the incorporation of the capped-profit subsidiary, OpenAI L.P.—which Musk alleges amounted to a “mov[e] to recast the [OpenAI] non-profit as a moneymaking endeavor to bring in shareholders, sell equity, and raise capital,” and an effort to “convert to a for-profit structure” — in a blog post in March 2019; Altman notified Musk of OpenAI, L.P.’s incorporation around that time; Musk was offered, and declined, equity in the new entity; and Musk raised no objection to the new entity’s formation.

Musk argued, no, it took longer than that:

Musk testified that he did not discover the truth about OpenAI’s full-blown commercial status until after Microsoft made its $10 billion investment in 2023. There was nothing unreasonable about that timing. OpenAI transformed into a commercial venture gradually, by degrees. Defendants sharply limited the information they disclosed and actively downplayed their commercial objectives in communications with Musk.

Musk argued that this was a question for the jury to decide, and yesterday it decided. Bloomberg News reports:

A jury rejected Elon Musk’s claims that OpenAI under Sam Altman’s leadership betrayed its mission to benefit the public by morphing into a for-profit business, finding that he waited too long to sue the company. ...

“I think there is a substantial amount of evidence to support the jury’s findings,” US District Judge Yvonne Gonzalez Rogers said when she accepted the nine-member jury’s unanimous conclusion after about two hours of deliberations. …

Musk and his lawyers vowed an appeal, but didn’t get into specifics about what they will argue.

“This reminds me of key moments in this country’s history, the Siege of Charleston, the Battle of Bunker Hill,” attorney Marc Toberoff said. “These were major losses for Americans, but who won the war? This one is not over.”

Sure okay. I suppose that the basic conclusion here is that OpenAI has not been a charity for many years, so it’s too late to complain that somebody stole it.

Hedge fund employees are very expensive. If you want to hire a very good portfolio manager, you might have to guarantee her like $50 million, and wait a year while she serves out her gardening leave. If you run a giant multistrategy hedge fund, you can afford that, but you need scale: You have to amortize the huge cost of employees over a lot of investor capital. If you run a smaller fund, you can’t really afford to hire the best portfolio managers, so your performance will suffer, so you will have a harder time raising money, so you will have an even harder time hiring, etc., in a vicious spiral.

If you do not run one of the giant multistrategy funds, it might occur to you to get around this problem with some sort of “Moneyball” approach. [3] Obviously the people who have established themselves as top portfolio managers cost $50 million and you can’t afford them. But they all came from somewhere; they weren’t making $50 million out of the womb. The trick, for you, is to try to find people who (1) are currently cheap but (2) can generate alpha. How do you do that?

Oh I mean there are lots of approaches involving, like, networking and interviewing and training promising candidates. But “giant internet open call for traders” might work too. Bloomberg’s Emily Nicolle reports:

Perpetuals.com Ltd. on Tuesday launched UpsideOnly, a platform that turns crowdsourced paper-trading strategies into shared real-money profits by executing bets with only the company’s capital. Investors place fake trades that predict the future price of assets like oil, gold and equities, after which an artificial intelligence model chooses the ones that are most likely to result in profit. The company then trades on those forecasts using its own money, and where it succeeds, shares half of the earnings with traders who created the winning signals.

In short, instead of parting with their cash, traders give up their time and expertise in training Perpetuals’ AI for a slice of the profits. And while the company itself takes on the risk of potential losses, it gains valuable data that AI could never generate on its own. ...

Crowdsourcing amateur trading signals to beat high-priced hedge fund strategies is a tried-and-tested model that hasn’t always worked out. Even with recent advances in the technology, AI has also proved to be poor at stock-picking or crypto trading when left to its own devices.

[Perpetuals CEO Patrick] Gruhn’s hope is that mixing the two might be a recipe for success. Perpetuals’ AI model, BayesShield AI, was developed in-house with the sole goal of scoring and identifying the strongest trading strategies, the company said. 

Basically a sort of distributed alpha capture approach. The trick here is that if you are recruiting professional portfolio managers, “we have some money and will give you half the upside of our trades with no downside” is not very appealing: Millennium is offering them half the upside of their trades on a much larger capital base and with a $50 million minimum. For amateurs, though, “half the upside with no downside” is a considerably better deal than getting the upside, and the downside, on their personal account.

US 30-Year Yield Hits Highest Since 2007 on Inflation Angst. The American Rebellion Against AI Is Gaining Steam. StanChart CEO Says AI to Replace ‘Lower-Value Human Capital.’ KPMG Taps Anthropic to Revamp Global Tax, Advisory Platforms. Distressed-Debt Deals Often End in ‘ Hard Default,’ Moody’s Warns. SpaceX IPO set to hand $20bn stake to one hedge fund. The Little-Known Hedge Fund That Stands to Make Over $10 Billion onSpaceX.Deutsche Bank fined by UK for breaching Russia sanctions. Treasury Lawyer Quits as Government Settles Trump IRS Suit. Judge Blocks Alex Jones’s Bid to Shield Infowars From Liquidation. Morgan Stanley-Led Group to Sell Chicago Parking Meters. The French mastermind behind a €1bn Ponzi scheme. (Earlier.)  Crypto Crime Escalates With Kidnappings, Cons and Human Coercion. Musk’s xAI Fails to Pay Staff $420 for Giving Their Tax Returns to Grok.

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!