I was reviewing a set of funding-rate histories for a colleague when the number first bothered me. Ten billion dollars. It appeared once in a November 2025 corporate summary โ Nvidia's private commitment to Anthropic, structured as strategic investment bundled with compute procurement. And then it appeared again, months later, attached to a different verb: anchoring the largest initial public offering in history. Same figure. Two stories. One of them had almost certainly borrowed the other's clothes.
That small duplication is the kind of detail a market trained on headlines never notices, and exactly the kind a risk auditor cannot walk past. Because around that single number, a quiet market has assembled itself. Nineteen crypto exchanges now list pre-IPO perpetual futures referencing Anthropic โ a company that has filed no prospectus, trades on no public venue, and has never confirmed the investment at the center of the trade. CoinGecko aggregates their implied valuations into an average of $1.94 trillion. Polymarket assigns a 90% probability to an event that has not happened. And somewhere in the middle of it, ordinary traders are pricing a trillion-dollar enterprise using an instrument with no underlying asset to which it can be tethered.
The story is not really about Anthropic. It is about what happens when crypto's most flexible instrument โ the perpetual swap โ is pointed at an asset that does not exist in any tradable form. Truth over hype. Always. But before we can say what that means, we have to be precise about what is actually being sold.
The Instrument Nobody Audited
A perpetual swap, at its core, is a bet that never expires. You hold a long or short position indefinitely, and the only thing that keeps the contract tethered to reality is the funding rate. When the perpetual trades above the spot index price, longs pay shorts; arbitrageurs step in, short the contract and buy the underlying, and the price is dragged back into line. That feedback loop is not a feature โ it is the entire safety mechanism. Everything else about perpetuals, including their enormous popularity, rests on the assumption that there is a spot market to arbitrage against.
Now remove the spot market. That is what a pre-IPO perpetual does. Anthropic's equity is not traded anywhere. There is no index of last-sale prices, no continuous quote, no venue where a share can be bought and delivered. When you strip out the anchor, the funding rate has nothing to pull against. It becomes decorative โ a number that computes faithfully and constrains nothing. The contract can drift arbitrarily far from any defensible valuation and stay there, because no arbitrageur can complete the loop. You cannot short the perpetual and buy the underlying when the underlying cannot be bought.
I have spent enough time auditing token distributions and reading whitepapers in bad lighting to recognize this pattern. In 2017, the flaw was always in the plumbing โ the part of the design that looked decorative until it was load-bearing. Here, the funding rate is the plumbing.
The Mark Price Is a Rumor Wearing a Number
With no spot price to reference, each exchange must construct its own reference. This is called the mark price, and in normal markets it is derived from multiple spot venues to resist manipulation. For pre-IPO equity, the exchange has no such luxury. It must either hire a private-market data provider โ Caplight or Forge spring to mind โ or simply assemble its own internal estimate from whatever private-round data it can gather.
Here is the consequence that almost nobody is discussing. Nineteen exchanges can use nineteen different mark-price methodologies, and none of them is obligated to disclose the source. A number that appears to be a market price may in fact be one desk's interpretation of a secondary transaction that may itself be months stale. The average implied valuation of $1.94 trillion is not a market price. It is an arithmetic mean across incomparable inputs โ a statistic with the visual authority of a market data point and none of the substance.
I have watched this pattern before. During the 2020 DeFi Summer, I wrote a series of long-form guides explaining how automated market makers worked, specifically for finance professionals who had never touched a wallet. The point I kept returning to was that a price is only as reliable as the mechanism that produces it. An AMM price is trustworthy because anyone can inspect the curve. An aggregated pre-IPO mark price is trustworthy only because a logo is attached to it. Noise filtered. Signal preserved โ but only if the filter is honest.
What the Ninety Percent Really Means
The prediction markets have been the loudest voice in the room. On Polymarket, the probability of an Anthropic IPO completing has moved from 67% at the end of October, to 85% by mid-November, to roughly 90% by year-end. Presented as a curve, it looks like conviction building. I read it differently.
Start with the base rate. A 67% starting probability for an event as operationally complex as a mega-cap IPO is not a cautious stance โ it is an optimistic one. Public offerings of that scale involve underwriters, auditors, exchanges, and regulators, any one of which can delay the process by quarters. To begin the market at two-in-three odds understates execution risk before a single line of a prospectus has been written.
Then look at the volume. Polymarket's cumulative traded volume on this narrative sits around $2.89 million. Set against a rumored $100 billion raise, that is a rounding error โ a few million dollars of noise attempting to price a hundred-billion-dollar event. The 90% figure is not the wisdom of crowds. It is the enthusiasm of a small crowd, amplified by a media cycle that finds '90%' a satisfying headline. When I see a probability curve rise that smoothly without matching volume to support it, I do not read it as information. I read it as anticipated self-reinforcement โ people betting that other people will keep betting.
The deeper issue is what the market is actually pricing. It is pricing the event of an IPO. It is not pricing the settlement of the derivative that references it, and those are not the same thing. A trader can be perfectly correct that Anthropic goes public and still lose money on a pre-IPO perpetual, because the contract's mark price, funding mechanics, and settlement terms are controlled by the exchange, not by the reality of the listing.
The Settlement Nobody Has Written Down
Perpetuals have no expiry, which is normally a feature. Here it is a trap. The economic event that gives the contract meaning โ the IPO โ is entirely exogenous, sitting outside the contract's control. So the exchange must define, in advance and in writing, what happens when the IPO occurs, and what happens if it never does.
Has any of the nineteen exchanges published those terms in a form a retail trader can verify? Based on what is public, the answer is essentially no, or at best inconsistent. This means holders face a settlement event that is defined by a counterparty with an incentive to minimize its own exposure. If the IPO is delayed, does the position roll indefinitely? If the IPO is priced differently than the mark price assumed, how is the gap resolved? If the exchange decides the reference event has 'materially' occurred, who adjudicates that word?
I have seen this movie. In 2022, when the market collapsed and platforms faced positions they could not honor, the fine print became the whole story. That year I restructured my own team's coverage to focus on fundamental resilience rather than speculative advice, and I watched three junior analysts learn, painfully, that the difference between a stated risk and an enforceable one is usually buried in a terms-of-service document nobody reads until it is too late. The most dangerous risk is not the one you can see on a chart. It is the one that only becomes real at the moment of settlement, when the other side is writing the rules.
Liquidity, Leverage, and the Manipulation Surface
Set aside the settlement question and look at the order book. A perpetual contract with no spot anchor and thin liquidity is a gift to anyone with size. Perpetuals typically support leverage of five to fifty times, and with no arbitrage force to punish a wrong price, a single well-capitalized participant can move the implied valuation by posting aggressively into a shallow book.
This matters because the $1.94 trillion figure itself may be a product of this dynamic. If the aggregate is built from venue-level marks, and those marks are set in thin markets, then the headline number is not a discovery โ it is a narrative price written by whichever wallet was willing to push hardest. The irony is sharp: a market ostensibly created to let people price a private company's future is, in practice, priced by people betting on what other people will believe about that company's future.
There is a broader pattern here worth naming. In my work covering cross-chain systems, I have argued the same thing for years โ that a bridge holding billions is only as safe as its most obscure component, and that cumulative bridge losses past $2.5 billion prove the industry keeps trusting bridges that its own audits warn against. The pre-IPO perpetual is that same faith, one layer removed. It is a bridge between a private-market asset and a public speculative market, and like the earlier bridges, the risk does not live in the parts that are working. It lives in the anchor that is missing.
The Regulatory Room Nobody Mentioned
Run the structure through the Howey test and the result is uncomfortable. There is money invested โ the margin. There is a common enterprise โ nineteen venues referencing the same company. There is an expectation of profit โ that is the entire premise. And that profit depends on the efforts of others โ Anthropic's management and Nvidia's capital. All four prongs lean the same way.
That points toward a legal characterization few of these products want: a synthetic equity derivative that looks and behaves like a security-based swap. In the United States, offering such an instrument to retail users without registration raises serious questions, and the questions come from two directions at once. The SEC looks at whether this is an unregistered security. The CFTC looks at whether it is an unlawful retail derivative or event contract. Polymarket's own history with the CFTC should have been a warning shot that this terrain is contested.
The likely reason all of this exists is that it does not sit in the United States at all. The legal wrapper is probably offshore, offered through international entities that deliberately sit outside the reach of the agencies most likely to object. Which produces the strangest feature of the whole structure: the underlying transaction โ if Nvidia really is investing in Anthropic โ is entirely lawful and unremarkable. It is the crypto derivative built on top of it that lives in the grey. A compliant foundation, feeding a channel that may not be compliant at all.
The Circular Foundation and Nvidia's Three Hats
And now to the thing the headline writers keep stepping around. Look at what Nvidia actually does in this ecosystem.
First, Nvidia sells the chips. Second, Nvidia stands behind the data center lease obligations that let those chips be deployed at scale โ commitments reported to run into the neighborhood of $105 billion. Third, if the reports are to be believed, Nvidia is an anchor investor in the very company that will spend heavily on those chips. Those are three distinct roles held by a single actor, and they form something that looks less like an ecosystem and more like a loop: invest in the customer, sell to the customer, guarantee the infrastructure the customer rents, and watch all three positions rise together.
This is the circular-financing debate in miniature, and it deserves to be handled with the calm precision this industry so often lacks. I am not alleging anything improper. Every leg of that loop can be individually sound, and the underlying demand for AI compute may be entirely genuine. But a market that prices its valuations from this loop should understand that it is not pricing an independent cash flow. It is pricing a financial reflex, and reflexes reverse suddenly. When I built the regulatory-literacy work my team runs โ translating frameworks like MiCA into material ordinary investors can actually use โ the recurring lesson was this: the number on the screen tells you what someone is willing to pay, not what something is worth. A loop can hold a valuation up longer than fundamentals alone, and it can let it fall much faster.
That $105 billion guarantee is the specific thread worth pulling. Sizeable lease guarantees are contingent liabilities, and contingent liabilities have a habit of appearing on balance sheets at the worst possible moment. If compute spending slows, the guarantees that support the valuation of the entire AI complex come into question โ and the pre-IPO perpetuals, sitting at the speculative edge of that complex, would be the fastest-moving instrument in the system. High-beta proxies are always the first to fall, and these instruments are high beta by construction, because they have no floor beneath them except a mark price someone else chose.
The Contrarian Reading: The IPO Is Not the Risk
Here is where I expect to lose the room.
Everyone is debating whether Anthropic can beat SpaceX's listing and how large the raise will be. I think that debate is almost irrelevant to whether a trader using these perpetuals makes or loses money. Consider the two ways to be wrong here. If the IPO never happens, the contracts may simply be adjudicated by the venues that wrote them, with the terms favoring the venues. If the IPO happens exactly as the bullish case predicts, traders can still lose โ through funding payments that run against an unpinned rate, through a mark price that never quite converges to the listing price, or through a settlement decision that resolves a fraction of the gap.
The genuinely contrarian point is this: the derivative can extract losses from participants who are directionally right. That is the opposite of how a well-built instrument behaves. A good derivative transmits your view faithfully; a bad one separates your view from your outcome. When the anchor is missing and the settlement is discretionary, being correct about Anthropic tells you nothing about being correct about the trade. Most people reading the headlines about this story are analyzing the wrong object. They are studying the company. They should be studying the contract.
There is a second, quieter blind spot. The narrative that crypto is finally 'democratizing' access to private-market upside is not new โ it is the same pitch I have heard at every cycle, from ICOs to yield farming to this. Each time, the access is real and the protection is not. The order of operations matters: access without a trustworthy pricing mechanism is not democratization. It is the sale of a ticket to a game whose rules are still being written by the house. Trust is the only currency that matters, and here the currency is being minted by the same people doing the dealing.
Where This Leaves the Reader
The clearest beneficiary of all this is not a trader. It is the exchange. A new, low-barrier product that generates fees and attention can be deployed across nineteen venues within weeks โ which tells you how thin the technical moat is. When I audited ICO structures in 2017 and found three separate token-distribution flaws that risked centralization, the lesson was the same then as it is now: when a product spreads faster than it can be understood, the margin of safety has usually been traded away for the margin of speed.
If there is a durable opportunity buried in this story, it is not in the perpetual. It is one layer underneath โ the private-equity valuation data that must be pulled on-chain to give these products any handle at all. Whoever builds a transparent, audited, credibly neutral reference for private-market pricing solves the actual bottleneck, and it is the least glamorous part of the whole narrative. The shovel, as usual, sits quietly while the gold rush shouts.
I will be watching one signal more than any other: whether the volume ever matches the probability. Ninety percent built on a few million dollars of trades is a claim, not a conviction. And the distance between the two โ between what the market says and what it has actually paid to say it โ is exactly where a careful reader should stand. So watch the settlement terms, not the headline. Watch the mark price, not the average. And ask, every time the number appears twice, whether the first one was real and the second one was borrowed. Noise filtered. Signal preserved. The signal here is not Anthropic's valuation. It is that we are once again pricing the invisible with an instrument built for the visible.
