What is the best way to get economic exposure to Anthropic stock right now? There are bad answers (expensive SPVs, fake SPVs). Here are two arguably good answers:

1. Just own Anthropic stock.
2. Own perpetual futures on Anthropic stock on Hyperliquid.

The first answer is pretty self-explanatory. Anthropic has stock, and people own it, and owning it gives you economic exposure to Anthropic stock. There are problems. Not just anyone can own Anthropic stock. It is a private company, and its shares do not trade on an exchange. You more or less need to get the stock from Anthropic, so this approach requires you to be an employee or a venture capital investor in one of its funding rounds or an effective altruist hedge fund manager. For most of us, that rules out this approach entirely. But there is another problem, even if you can get the stock. The other problem is that Anthropic stock is fairly illiquid. [1] You can’t easily sell it, until it goes public. Owning the stock gives you economic exposure to the stock, but you don’t get to decide how long you keep that exposure. And you don’t really know its price. So you do have economic exposure to the stock — if it goes up, your stock is worth more — but you don’t necessarily know how much.

The second answer is weirder. Hyperliquid Labs runs a leading platform for trading perpetual futures contracts. Traditionally perpetual futures are a crypto thing, a way to make levered bets on cryptocurrencies, they have expanded into stocks and commodities. Kalshi launched an S&P 500 futures contract and is working on perpetual futures for individual stocks. Perpetual futures contracts are basically bets on prices: One person goes long, one person goes short, the short pays the long if the stock goes up, and the long pays the short if it goes down. A perpetual futures contract on an individual public stock is just a levered bet on that stock, and by arbitrage it will closely track the price of the underlying stock. If SpaceX stock trades at $167.60, and SpaceX perpetual futures trade at $165, arbitrageurs will buy the futures and sell the stock until the prices are in line, etc. [2]

But perpetual futures on private stock have no arbitrage mechanism to make the price correct. If Anthropic perpetual futures trade at $120, while Anthropic’s stock is at $100, arbitrageurs cannot sell the futures and buy the stock until the prices converge: They can sell the futures, but they can’t buy the stock. [3] Also, how do you even know that “Anthropic’s stock is at $100”? Anthropic doesn’t trade liquidly or publicly, so you don’t know its price from minute to minute, the way you know the price of the futures. [4]

Also, these perps are pretty obviously illegal in the US: Under US securities law, you can’t sell bets on a company’s stock unless that company has registered its stock with the Securities and Exchange Commission (i.e., gone public). So these are a non-US product. Coinbase offers “pre-IPO perpetual futures,” but only outside of the US. Polymarket offers bets on private companies’ stocks, but again only outside the US, heh heh heh. Robinhood too. And Hyperliquid “blocks US users as per industry practice, but anyone can trade on the blockchain that powers it.” [5]  

So an Anthropic perpetual is a bet, on a weird offshore crypto betting platform, on the price of Anthropic stock. This does not sound like a great way to get economic exposure to Anthropic stock. It’s just that, empirically, it kind of is? We talked about Hyperliquid pre-IPO perpetual futures on SpaceX back in June, when SpaceX was going public; the perps did a good job of predicting the first-day price of SpaceX. I quoted TD Securities analyst Reid Noch, who described perpetuals like this:

A pre-IPO PERP has no public stock price to reference because the company has not listed yet. Instead, the market prices themselves are solely based on supply and demand. Once the stock begins trading, the contract can then anchor to the listed equity price. Until that point, it acts more like a leveraged, continuous prediction market than a substitute for owning the stock.

SpaceX was not Hyperliquid’s first success; Noch cited another pre-IPO perp, on Cerebras:

Hyperliquid largely predicted the post-IPO pop without offering true exposure to the underlying stock and without a public reference price to anchor the contract. Traders did not own Cerebras shares, receive IPO allocation rights or have any claim against the issuer. They were trading a cash-settled derivative on where the market believed the stock would open.

This is a little unsettling. It is not obvious that bets on an offshore crypto platform would give you any economic exposure to Anthropic. There’s no, like, theoretical reason for those bets to be right. They could be way off. It’s just that often they are right.

So that’s not great. But in other ways the perps are better than owning the stock:

1. Anyone can buy perps. I mean, not anyone; they are technically off limits in the US. But in crypto that tends to be a pretty soft suggestion. (Not legal advice!) And you don’t need Anthropic’s permission.
2. They offer constant liquidity and price discovery. You always know the price of your perps, and you can sell them any time.

And so it is plausible that, for at least some investors, the actual best way to get economic exposure to Anthropic stock is on Hyperliquid. (Not investing advice!) Like, the ranking might be Hyperliquid perps first, actual Anthropic stock second. I feel like people will be mad at me for saying this, and they will be correct. I am mad at me for saying this. It shouldn’t work. Obviously once Anthropic goes public, owning the stock is better. [6]  But for now, the perps have some appeal.

In particular, if you are an exchange-traded fund, perps might be better than owning actual private-company stock, for several reasons:

1. ETFs trade on the stock exchange. The price of an ETF share tracks its net asset value per share. An arbitrage mechanism (called “creation/redemption”) ensures this: If the NAV is $20 per share, and the ETF is trading at $21, arbitrageurs can buy a basket of the underlying assets for $20, deliver it to the ETF, get back an ETF share and sell it for $21. This works great when the ETF is a basket of public stocks: Arbitrageurs can easily trade the underlying stocks against the ETF to make sure that the prices stay in line. It works less well when the ETF is a basket of private stocks, since arbitrageurs can’t trade those. But you can trade perps all day long, no problem. Deliver some perps, get back ETF shares, done.
2. Similarly, the ETF will want to report its net asset value per share. If it owns private stock, it may have a hard time getting a reliable price on that stock from day to day or minute to minute. If it owns perps, easy peasy.
3. There is a classic “liquidity mismatch” worry with ETFs: ETF investors can take their money out at any time, which is bad if the ETF’s assets are less liquid than its (very liquid) shares. If an ETF owned a ton of Anthropic, and everyone asked for their money back, it would be a mess: How could the ETF sell the illiquid restricted Anthropic shares? [7]  The perps are more liquid.

Of course most ETFs do not even aspire to hold Anthropic stock; they track a public stock index, not private companies. But if you did build an ETF that held Anthropic, there would be demand. [8] And if you were looking to do that, you might want to build it with perpetual futures rather than with actual Anthropic stock. Sure sure sure, your ETF wouldn’t actually have any Anthropic stock, and there’s a risk that the perps would end up being a poor proxy for that stock. But, for now, the perps have nicer technical properties than the actual stock.

Behold MN, the Corgi MANGOS ETF:

Concentrated actively managed exchange-traded fund that offers investors exposure to OpenAI and Anthropic, two private companies in artificial intelligence, alongside Meta Platforms, NVIDIA, Alphabet (Google), and SpaceX. MANGOS is an acronym formed from the first letter of each of the six companies' names.

It started trading last week. It closed yesterday at $26.12, versus a reported net asset value of $26.26. It reports $68.4 million of assets under management, 5.25% of it in Anthropic and 5.12% in OpenAI. (The rest is in big public AI companies, plus some cash.) Where does it get its Anthropic and OpenAI? Oh yes:

For privately held companies, including OpenAI and Anthropic, the Fund expects to use swaps and/or eligible SPVs rather than direct purchases of private-company shares. Following a public listing, the Fund may use listed shares and financial instruments, including swaps. ...

Private-company swaps are expected to use reference pricing derived from perpetual futures contracts referencing the applicable company. ... A perpetual futures price may differ materially from the value of the company’s shares. The Fund does not itself hold perpetual futures or maintain an account with, or post collateral to, a perpetual futures venue under this strategy.

“At the moment,” tweeted Corgi Invest Chief Investment Officer Jeff Weniger, “the exposures to those two are via total return swaps on the perpetual futures,” [9] and the replies are extremely enjoyable. Economically, the ETF owns perpetual futures on OpenAI and Anthropic, but in practice US ETFs tend not to trade on offshore crypto betting platforms directly, so instead Corgi enters into total return swaps on the perps — a further bilateral bet — with some less regulatorily constrained trading firm. We have discussed a proposed sports gambling ETF which would do the same thing: It would make sports bets on Kalshi, but for legal purposes it would do those trades on swap through an intermediary. The ETF holds a normal-ish financial instrument, an over-the-counter swap with a bank or trading firm. The swap counterparty owns the weird stuff.

Should a US listed ETF, available to retail investors, contain bets on private-company stocks made on weird offshore betting platforms? Well! I mean! Have I mentioned the sports gambling ETF? 

I … find myself hating this less than I should? This is … financial innovation? We have discussed Corgi before, because it launches lots of oddball active ETFs; “the spaghetti cannon is firing wildly right now,” someone once said. This one is a little insane, yes, but I see how they got there.

Nothing in this column is ever investing advice, especially not any of this. If you come to me and say “I bought that MANGOS ETF because you told me to” I will smack you. I did no such thing! I just appreciate comedy in financial engineering, and we’re getting a lot of it these days.

A big theme around here is that banks are re-tranching. In the olden days, banks did some risky economic activity. In 2026, banks probably do (1) less of that risky economic activity but (2) more lending to other companies that do that activity. The banks have moved up in seniority: Instead of direct exposure to some economic risk, they have some senior claim on some other company that takes the economic risk. If the risk goes bad, the other company takes the first loss; the bank only loses money if the risk goes really bad.

A lot of this is about private credit: In the olden days, banks would make loans to companies and credit-card customers and what have you; now, private credit funds make some of those loans, but the banks make loans to the private credit funds. Instead of a bank lending $100 to companies, the bank will lend $50 to a private credit fund, which will use that $50 and $50 of its investors’ money to make $100 of loans to companies, etc. 

But a lot of it is about trading. Twenty years ago, the “trading” business of a big investment bank mostly did trading: It would buy and sell stocks and bonds and derivatives, making money by buying low and selling high. Now, the “trading” business is increasingly a lending business: The bank lends money to hedge funds and proprietary trading firms, so that they can buy and sell stocks and bonds and derivatives. And the hedge funds and proprietary trading firms increasingly fill the economic roles that used to be occupied by banks. A couple of years ago, I wrote:

The simple, crude, overstated model is: In 2004, a bank’s trading division was in the business of trading with the bank’s money. In 2024, a bank’s trading division is increasingly in the business of lending the bank’s money to hedge funds, which do the trading.

That keeps being more true. The Financial Times reports on the continuing rise of prime brokerage:

US banks are generating record trading profits. But they are not the ones doing the trading.

Firms such as Jane Street and Citadel have aggressively expanded into banks’ traditional turf, dominating the core business of market making as well as placing big bets on market moves. But their insatiable demand for financing to juice those trades has made them irresistible clients for lenders and a powerful engine for profits.

In market making, “banks are struggling to compete”, said Mike Webb, Barclays’ global head of liquid financing. “But financing is a thing that banks do and there is a big moat around it.”

The moat is built on credit lines that lenders provide to trading giants that run into trillions of dollars. Hedge funds’ borrowing from banks has tripled since 2020, and the revenues from financing have turned prime brokerage into one of Wall Street’s fastest-growing businesses. …

Much of the speculative trading that once used to take place on banks’ prop desks migrated to the hedge fund industry, while market making — a lower-risk activity — became the purview of high-frequency trading firms and later specialist market makers. …

Prime services revenues will have leapt from 10 per cent of banks’ equities revenues in 2005 to about 38 per cent this year, according to Coalition Greenwich — and those figures exclude the vast derivatives services banks provide.

The basic tradeoff with all of this re-tranching stuff is that the banks get a more senior claim on the economic activity, but they have less control over it. A bank that owns a bunch of stocks and bonds has more direct market risk than one that just lends a lot of money to Jane Street, but a bank that lends a lot of money to Jane Street has less ability to manage that market risk. To some extent you have to trust Jane Street:

Because the managers are sophisticated players with hundreds of trading teams across a spectrum of asset classes, they have an inbuilt diversification and level of risk management that has allowed banks to justify ratcheting up their lending.

Webb at Barclays said: “An institutional-quality hedge fund or asset manager is a sophisticated organisation with a liquidity-management function that reports directly to the founder.”

One other point about this re-tranching. I have often argued that the rise of private credit basically makes the financial system safer. Banking is inherently risky: Banks get a lot of their funding from deposits, which could be withdrawn at any time, and use those deposits to make long-term loans. Private credit funds are a safer form of narrow banking: They raise long-term locked-up equity financing from investors (mostly!), they take out term loans from banks, so they have less funding risk. A migration of lending activity from banks to private credit probably makes the financial system safer, though there are counterarguments. 

Traditionally you would not have said the same thing about prime brokerage. Traditionally, hedge funds and proprietary trading firms got a lot of their financing from their prime brokers in quite short-term forms: Banks gave them margin loans that had to be repaid if they lost money, and the prime brokers tended to have a lot of flexibility to withdraw financing, change margin terms or demand more collateral if they got nervous. A migration of trading activity from banks to hedge funds would not make the financial system any safer. The banks’ funding model — with loyal depositors, with access to the lender of last resort, etc. — was probably more robust than the hedge funds’ model.

But that’s shifting a bit, the FT notes:

The growing scale of a few clients at the top of the industry has also shifted the balance of power, according to people familiar with industry practices, with clients pushing for far more favourable lending terms.

Lock-up terms, the length of time a prime broker commits to providing financing before it can withdraw credit, have gradually lengthened.

Commitments that typically lasted two weeks went up to 30 days, stretched to three months and more recently to half a year for some of the industry’s most coveted clients. That means a bank cannot cut back its financing or demand more margin than agreed at the start of the contract without giving increasingly lengthy notice.

If big hedge funds and proprietary trading firms are systemically important traders, they need to have some long-term locked-up funding.

We have talked a few times about systematic investing strategies in prediction markets. They all have this basic thesis:

1. People like a gamble. That is why they are on prediction markets.
2. Betting on longshots is more fun — more of a gamble — than betting on sure things. It’s more fun to bet a little bit of money to have a small chance to win a lot of money than it is to bet a lot of money to have a large chance to make a little bit more. It’s more fun to bet on the underdog than the favorite in football. It’s more fun to bet on the imminent return of Jesus Christ than it is to bet that the next year will, eschatalogically speaking, resemble the previous 2,000 years.
3. Therefore, the gamblers on prediction markets will predictably favor longshots. They will buy more of longshot contracts, pushing up the prices (and market-implied probabilities) of those contracts. If something has a 2% true probability of happening, people will buy it, so the Yes contract will trade at like 5 cents on the dollar and the No contract will trade at like 95.
4. Therefore, if you systematically sell longshot contracts, you will predictably make money. Sometimes you’ll lose money — things with 2% probabilities happen 2% of the time — but those times are infrequent and basically uncorrelated. You can buy 100 different No contracts for $95; two of them (2%) will lose, but the other 98 will pay off at $1, and you’ll make $3.
5. Perhaps this is all rational, and you are getting paid a sensible insurance premium for taking on some sort of tail-event risk. (Maybe all the correlations go to one if Jesus Christ does return.) Or perhaps — surely? — this is a behavioral bias of actual prediction-market participants, and you are getting paid for taking the other side of mass gambler irrationality.

We have discussed various forms of this basic strategy, including:

• Selling sports parlays;
• Betting on extreme favorites in sporting events;
• Betting that Donald Trump-related events won’t happen; or 
• Just betting that events in general won’t happen. 

In each case, someone has found evidence that the strategy works. The meta-strategy of “prediction market gamblers overestimate the probability of weird stuff, so bet against weird stuff” is psychologically plausible and also often works. Not investing (or whatever this is) advice.

Bloomberg’s Carolyn Silverman and Denitsa Tsekova provide more evidence:

Pore through the reams of bets made on Polymarket and Kalshi and a clear pattern emerges: bettors flocking again and again to money-losing longshot wagers.

To this crowd, the sites aren’t the truth machines that their billionaire founders claim, converting millions of carefully considered wagers into precise forecasts of real-world events that will, in turn, inform decision-making by businesses and investors across the globe. No, for many customers, these markets look more like a casino crossed with an arcade, open 24/7 and capable of serving up a never-ending buffet of nearly impossible bets that maximize the dopamine rush even as they minimize the financial returns.

Since the start of 2025, around half of all trading volume on Kalshi, the largest prediction market platform, has involved one side chasing potential returns of at least 1,000% betting against mostly sophisticated traders and market-making firms on the other side, a Bloomberg News analysis of over a billion transactions shows. In recent weeks, that percentage has climbed even higher, up to more than 65%, as bettors piled into sports combination wagers, known as parlays, in pursuit of big scores.

On Polymarket International, which, unlike the firm’s US entity, has a public ledger of trading data, the figures are similar. Those trades on unlikely outcomes — such as the likelihood of Jesus Christ returning to Earth or LeBron James winning the US presidential election — accounted for 52% of all volume over the past 21 months. …

For the firms and traders willing to stake the capital and offer up those long odds, the wagers have churned out a flow of steady, albeit modest, profits. For the bettors taking those odds, though, the losses are piling up. Sure, they hit a 40-1 shot every once in a while, as on traditional sportsbooks, but not at nearly a high enough rate to offset all the losing bets they make.

Yes. Look. Obviously. Prediction markets are casinos. The interesting thing about them is that they are all-to-all casinos: For every trader who wants a fun gamble, there is a (probably better capitalized) trader on the other side offering the fun gamble. Some people are buying fun, and other people are selling fun for money.

Elsewhere: “NFL tells Supreme Court prediction markets are gambling, not investing.” Yes everybody knows this! A lot of people pretend not to.

A reader sent me this Instagram Reel/YouTube short from a Malaysian food influencer who claims he bought 100 shares of stock of a public company, for about $18 total, in order to get access to its annual shareholder meeting, which serves a free buffet. “You buy share for profit, I buy share for huge chunk of meat, we are not the same,” he says. The stock is down about 3% since the annual meeting, so the lunch cost him about 50 cents. He says the food isn’t that good. I don’t know that this is a great trade but it’s a funny video.

Goldman’s Top Leaders Set to Score More Than $500 Million With Special Bonus. When 20% of Profits Won’t Do, Fund Managers Seek ‘Super Carry.’ Oracle, Broadcom and SpaceX Seek Blockbuster Debt Deals to Pay for AI Chips. Tiger Global Poised for $5 Billion Windfall From OpenAI Bet. SpaceX credit risk jumps on worries over its borrowing spree. Nvidia-Backed IPO’s Cratering Demand Sends Warning on AI Funding. The AI Price War Is Heating Up — and OpenAI Is Gaining Ground on Anthropic. Partners Group Gets Creative to Fix Funds Hit by Redemptions. Wells Fargo Faces Regulator Probe Over Efforts to Boost Black Homeownership. Big investors ‘bottom fish’ in Eurozone bond markets after France sell-off. Ellisons Invest About $17 Billion in Warner Bros. Takeover. Miami Developer Gains $1 Billion Windfall From Griffin’s Real Estate Buy. Wise to pay customers’ tax bills after errors. AI upends Singapore’s ‘quant Olympics.’ AI agent spam phone calls.

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!