On Tuesday, a derivatives venue called MNX went live on MegaETH, and its opening order book told a colder story than its press release did. The contracts it lists — futures tied to the private valuations of companies like OpenAI, Anthropic, DeepSeek, and Moonshot — have no underlying spot market, no external oracle, and no continuous price feed. The settlement price is not read from the world. It is manufactured inside the venue by the same traders betting on it. Logic does not bleed, but code leaves traces, and the trace here is a loop: the market prices the asset, and the asset is the market's own opinion.
I have spent enough hours in contract explorers to distrust any product whose price engine is also its audience. When I audited the yield aggregator that drained $30 million in 2020, the fatal detail was not a missing check — it was a data source the developers trusted because it felt authoritative. MNX has done something structurally similar, only inverted. It has removed the oracle entirely, and in doing so removed the thing that would have anchored its price to reality. That is not a security improvement. It is a substitution of one trust assumption for a larger one.
Context: What Manifold Actually Built
MNX is the product of Manifold Markets, a prediction-market platform with a real delivery record. That pedigree matters. Manifold's core competence has never been blockchain throughput; it is the design of mechanisms that assign prices to events that have no natural market — elections, court rulings, whether a specific founder will tweet before noon. The migration from "price the future event" to "price the private company" is not a genre jump. It is the same muscle, flexed at a larger weight. Stephen Gruggett, a Manifold co-founder, has spoken publicly about why the venue uses dated futures rather than perpetuals: private companies do not trade continuously, so a contract without an expiry has no natural moment to settle at truth.
That reasoning is sound, and I want to credit it before I take it apart. The venue raised $6.4 million in June, led by Village Global, a network-oriented early-stage fund. It is building on MegaETH, a young high-performance L2, rather than Ethereum mainnet or a mature rollup. It states plainly that it will not serve US customers and plans to acquire the necessary licenses. It caps leverage at 3x and charges no periodic funding rate.
Every one of those facts is a design decision, and each one can be read two ways. That duality is the entire story.
Core: The Anatomy of a Price That Has No Source
The most counter-intuitive element of MNX is not that it lists private companies. It is that the price of those listings is endogenous — determined by trading activity inside the platform, not by any external reference. Strip the marketing and this is a prediction market wearing a futures contract's clothes. The venue is asking its own users to collectively guess what Anthropic is worth, and then treating the median guess as a price.
This solves a genuine problem. Private companies have no continuous price, which is precisely why pre-IPO derivatives have historically been a graveyard. Forge, EquityZen, and their peers move real shares, slowly and expensively. MNX's answer is to skip the shares entirely and let capital inside the venue discover the number.
But a discovered number is only as good as the liquidity and rationality behind it. Consider the mechanics under stress. If daily volume in the OpenAI contract is thin — and for a brand-new venue listing illiquid private assets, thin is the base case — then the quoted price reflects the beliefs of a handful of wallets, not the consensus of a market. A single well-capitalized participant can move the mark meaningfully. This is not a theoretical manipulation vector; it is the default state of any market that opens without a natural market-maker cohort. Private-company derivatives have no natural market-makers. There is no arbitrageur who can hedge an Anthropic futures position with a correlated instrument, because no such instrument exists. The usual mechanism that pulls a price toward truth — the arbitrage that links a derivative to its underlying — is simply absent.
What remains is a self-referencing system. The price is the crowd's belief, and the crowd's belief is the price. Feedback loops of this shape are not markets in the classical sense. They are coordination games, and coordination games settle wherever the loudest participants decide to settle.
The settlement mechanism is where the structure either holds or cracks, and it is the weakest joint in the frame. At expiry, if the underlying company has not gone public, the contract resolves against a valuation drawn from a funding round, a fund mark, or a secondary-market print. Read that list again. A funding round is discrete, infrequent, and often a year stale. A fund mark is an opinion from a single general partner. A secondary print is whatever one seller and one buyer agreed to on a given Tuesday. None of these is a continuous, verifiable, tamper-evident price. The gap between what the market expects at expiry and what the settlement source actually reports is a basis risk large enough to swallow the product. Volume is noise; the wallet cluster is signal — and at settlement, there is no cluster to read, only a document someone chose to honor.
Now watch the leverage. Three times, no more. A venue that genuinely believed in its own price discovery would allow more. The cap is an admission that the founders themselves are uncertain how far the mark can drift from reality. It is prudent, and prudence is not the same as confidence.
There is a second architectural choice worth naming. A venue that refuses US customers and simultaneously announces a licensing roadmap cannot be running permissionless DeFi. KYC and jurisdictional gating require identity, and identity requires a controlled front door. The realistic architecture is centralized matching with on-chain settlement — a hybrid, with an operator who can see, gate, and in principle intervene. That is not a scandal. It is simply the truth that the "decentralized exchange" label cannot survive contact with a securities-adjacent product that must know who its users are.
And the chain choice is instructive. Building on MegaETH rather than mainnet buys performance and cost, and it buys something else: a young L2 desperate for a flagship application will offer terms, attention, and ecosystem support to an early tenant. The relationship is bidirectional marketing. MNX borrows MegaETH's throughput narrative; MegaETH borrows MNX's "first serious app" credential. Both are real benefits. Neither reduces the risk that the underlying chain is itself early and unproven under load.
I ran the same lens over a top-tier PFP collection in 2021 that claimed a billion-dollar market cap. I scraped three months of transfers and found that roughly 60% of the volume traced back to a single entity washing its own floor. The lesson was not that the collection was fake. It was that a price with no independent anchor will report whatever the largest participant wants it to report. MNX's private-company contracts have the same structural property, and they have it by design.
The Regulatory Fault Line Runs Through the Product
Apply the Howey test to a futures contract on a private company's equity and the answer comes back uncomfortable. There is money invested. There is a common enterprise — every trader on the same contract shares exposure to one company's fate. There is an expectation of profit. The only element in genuine dispute is whether that profit derives from the efforts of others, and when the underlying is a private operating company run by a management team, the honest answer is: partly yes. A derivative whose reference asset is a security is, in the US framework, a security-based swap — which pulls it toward SEC jurisdiction even as the futures wrapper points toward the CFTC. The product sits in the narrowest, most contested seam of financial regulation.
"We do not serve US customers" is a clean short-term answer. It is also a voluntary surrender of the largest pool of capital and the largest cluster of AI companies on earth. A venue that wants to scale cannot permanently exile itself from the market where its underlying assets actually live. The licensing plan acknowledges this. It just has not solved it.
Layer on the nationality of the underlyings. DeepSeek and Moonshot are Chinese companies; OpenAI and Anthropic are American. A single venue now straddles cross-border data rules, currency controls, and geopolitical sensitivity on two continents at once. The compliance surface is not a wall. It is a lattice, and every gap in it is a jurisdiction that can shut the door.
There is a quieter legal exposure that deserves naming. When a venue settles a contract against a private company's funding-round valuation, it may be using non-public information about that company without consent. The company being priced never agreed to be priced. If Anthropic or OpenAI decides it does not want its valuation traded speculatively by strangers, it can pressure data providers, threaten legal action, or simply stop sharing numbers. The product depends on the cooperation — or at least the indifference — of entities that have every incentive to withhold both.
The Contrarian Case: The Bulls Are Not Wrong About the Gap
I have spent this article dissecting the product's architecture, and it would be dishonest to stop there, because the bulls are correct about something important. The gap MNX is trying to fill is real. AI is one of the largest capital-formation cycles in modern history, and the financial infrastructure around it is genuinely primitive. Public-market investors can buy Nvidia; they cannot buy a private AI lab's progress. That asymmetry is not a narrative. It is a market failure, and market failures attract builders.
Manifold is also a credible builder for this specific failure. The team has spent years pricing events that had no price. Extending that capability from "will X happen" to "what is Y worth" is a natural, defensible progression, not a pivot into unfamiliar territory. The named founder, the clear investment thesis articulated by Max Killberg — build financial markets around AI rather than betting on yet another prediction market — these are the signals of a team that understands its own edge.
The dated-futures design is likewise more thoughtful than the perpetuals offered by larger exchanges. A perpetual has no expiry, which means it has no moment of settlement, which means it can only sensibly track an asset that already trades continuously. For a private company, that is a category error. MNX's choice to use an expiry is the correct one, and it reveals that the team has thought harder about the underlying than the venues it is being compared to.
Where the bull case quietly breaks is in assuming that a real problem implies a real product. The pre-IPO secondary market has existed for decades and has never scaled to the size its logic implies, precisely because private-asset liquidity is structurally thin and settlement is structurally subjective. MNX has not dissolved those constraints. It has wrapped them in an interface that loads faster. The demand may be real and the product may still fail, and those two facts are not in tension.
Takeaway: Watch the Settlement Document, Not the Launch Tweet
Imagination is infinite, but liquidity is finite, and the place where MNX will be decided is not its homepage. It is the settlement rulebook it has not yet published. Watch for three things over the next two quarters. First, the actual depth on the OpenAI and Anthropic contracts — if spreads stay wide and volume stays in the low five figures, the price is a rumor, not a market. Second, the disclosed valuation sources and their weighting, because a settlement mechanism with undisclosed discretion is a mechanism with a thumb on the scale. Third, whether any named underlying company publicly objects, because the first legal threat against a valuation will reveal whether the whole category can survive contact with the companies it prices.
The rug is not pulled; it was never tied. MNX has built something genuinely novel, and novelty in this space deserves scrutiny rather than applause. The question worth carrying forward is not whether private AI companies should have a market price. It is who gets to author that price, and what happens to the traders when the author changes their mind.