On Tuesday, MNX — a crypto exchange built by the prediction-market team at Manifold Markets — began listing futures contracts on private AI companies. Anthropic. DeepSeek. Moonshot. A fourth entity the source material renders only as OpenAI. None of these companies trade. None publishes a continuous price. None has agreed to be repackaged into a leveraged derivative and sold to strangers on a Layer 2.
That is the point. It is also the problem.
I have spent enough of my career reverse-engineering broken pricing mechanisms to recognize the shape of this one. In 2022 I paused three months of active trading to dissect how TerraUSD decoupled from LUNA — a system that, on paper, discovered price through an internal mechanism and, in practice, discovered only its own fragility. MNX is not that. But it belongs to the same family: a market that prices itself by looking inward.
Manifold Markets has run a prediction market for years — a venue where users wager on the outcome of future events. The methodological DNA is exact. Take a question with no live market price. Let participants bid against each other. Let the aggregate of their positions approximate a probability. MNX applies that DNA to a harder target: the valuations of private companies.
The mechanics matter. These are futures, not perpetuals, with expiry at the end of 2027. No periodic funding rate. Leverage capped at three times. The underlyings are unlisted — Anthropic and OpenAI on the US side, DeepSeek and Moonshot on the Chinese side. Pricing is set by trading activity inside the platform, not by an external oracle. Settlement, if a company has not gone public by expiry, references funding rounds, fund marks, or publicly reported secondary valuations. The financing is modest: $6.4 million raised in June, led by Village Global. Co-founder Stephen Gruggett attached his name publicly. Investor Max Killberg framed the thesis plainly — build financial markets around AI, the largest investment cycle in memory, rather than betting on another prediction market. MNX runs on MegaETH, a high-performance L2. It does not serve US customers, while stating an intention to obtain licenses elsewhere.
The pricing mechanism is a self-referential loop, and that is both its cleverness and its ceiling. Every pre-IPO derivative dies on the same rock: a private company has no continuous price. MNX sidesteps the rock by refusing to look for one. The contract price is whatever participants on the platform agree it is. This eliminates the oracle attack surface entirely — no external feed to manipulate, no data provider to bribe. But it replaces external manipulation risk with internal manipulation risk. The price becomes a mirror: it reflects the consensus of whoever is trading, not the fundamentals of the company traded.
For a market with deep liquidity and a diverse participant base, that mirror can inform. For a market that launched Tuesday with no natural market maker — and there is no natural market maker for a private AI company's 2027 valuation — the mirror mostly reflects how few people are in the room. Wide spreads. Thin depth. A price movable by a single sizeable order.
The futures structure, by contrast, is genuinely well-reasoned. OKX and Bitget, the competitors the source material cites, offer perpetuals. A perpetual is fine for an asset with a continuous price — BTC, ETH, a liquid token. It is the wrong instrument for an asset whose true value crystallizes only at discrete moments: a funding round, an IPO, an acquisition. MNX anchors to an expiry because it needs a specific moment to settle real value. This is not cosmetic. It is the correct design for the asset class, and Gruggett's framing is technically sound.
But settlement is the Achilles' heel, and the source material is suspiciously quiet about it. The stated rule: if the company has not listed by expiry, the contract settles against a funding round, a fund mark, or a public secondary valuation. Consider what that means. Funding rounds are discrete, low-frequency, lagging — a company may reprice once a year. Secondary marks can lack a single authoritative source. Fund marks are opinions, marked quarterly and smoothed. Even if internal price discovery works perfectly during the contract's life, the settlement price can land far from where the market expected. That gap — basis risk between traded price and settlement reference — is not a bug to patch. It is structural. And the source discloses no valuation-source hierarchy, no weighting, no dispute-resolution mechanism. That silence is the most important data point in the story.
Survival is the ultimate metric of a robust system. A derivatives venue does not survive on clever pricing. It survives on settlement nobody can contest. MNX has engineered the first half and left the second half as an exercise for the reader.
There is no token, and that changes who this is for. Across every information point, not one mentions a token, an airdrop, or on-chain governance. This is a CeFi product funded by equity — $6.4 million, closed in June. Value accrues to shareholders, presumably through trading fees, but no fee schedule is disclosed. The implication for the reader is uncomfortable: the narrative is tradable in sentiment, not in instrument. There is no token to buy. The story reinforces the financialization-of-AI-assets theme without handing anyone a vehicle to express it. And $6.4 million is small for a team attempting a compliant derivatives exchange — compliance, market making, and engineering all cost money.
The MegaETH choice is a bet within a bet. Selecting a young high-performance L2 over Ethereum mainnet or an established rollup buys cost and throughput, and probably ecosystem support — new chains subsidize early flagship apps. It also imports the underlying chain's early-stage risk. If MegaETH stumbles, MNX inherits the stumble. It is a double endorsement: MNX borrows MegaETH's performance narrative; MegaETH borrows MNX's status as a marquee application.
The team carries methodological continuity, which lowers execution risk but not structural risk. Manifold has shipped real product in the prediction-market space; moving to private-company valuation futures is a natural extension of the same core capability — pricing events that lack a market. Gruggett's willingness to attach his name signals an appetite for reputational risk, a relative positive against anonymous teams. But governance is centralized. Decision transparency, custody of user funds, and dispute resolution are all undisclosed. Village Global is a reputable early-stage fund but not a Tier 1 crypto-native backer, and a $6.4 million round is a small base for a licensed derivatives venue. The investment logic is coherent; the capitalization is thin.
Regulation is the single largest risk, and it is not close. A futures contract referencing the equity value of a private company sits in the most contested intersection of financial law. Futures may fall under CFTC jurisdiction; a contract whose underlying is a security can trigger the SEC's security-based swap regime. Layer on the underlyings' nationalities — US names on one side, Chinese names on the other, with cross-border data and capital implications — and the compliance surface multiplies. The cost structure here echoes the pattern I watched in Europe under MiCA: broad clarity at the top, and an operating burden that quietly eliminates the small players. Not serving US customers is a clean short-term move. It also surrenders the deepest pool of capital and the highest concentration of AI companies on earth. A product that wants to scale must eventually solve the US question head-on.
Liquidity cold-start is the operational risk that compounds everything else. Private-company derivatives have no natural market-making community. There is no arbitrage desk with a hedge, no basis trade, no inventory model for a 2027 valuation of an unlisted firm. Launch volume will likely be thin, spreads wide, and the self-referential price unstable precisely when the venue needs it most. The three-times leverage cap reads as deliberate restraint — the platform knows it cannot support more — but restraint does not manufacture counterparties.
The contrarian read is that MNX is fighting the wrong competitor. The source benchmarks it against OKX and Bitget. But those venues are almost certainly listing perpetuals on publicly tradable AI proxies — tokens, tokenized equities, indices — not on private company valuations. They are not substitutes. MNX's real competitors sit outside crypto: Forge, EquityZen, the broader pre-IPO secondary market, and structured-product desks at banks. Those markets are illiquid and gated for a reason MNX does not remove — private companies do not want their valuations traded publicly, at high frequency, on speculation.
Which surfaces the blind spot: the underlying companies never agreed to this. If Anthropic or OpenAI objects, it can pressure data providers, threaten legal action, or simply stop disclosing anything. A derivative whose settlement references a valuation the reference company refuses to acknowledge is a contract with no reliable anchor. Benchmarking against centralized exchanges is flattering marketing, not accurate competitive analysis. The stress test that matters is not whether MNX can compete with OKX. It is what happens on the day the underlying refuses to cooperate.
The question is not whether MNX can price a private AI company. It is whether it can settle that price in a way no counterparty disputes. Watch two numbers after launch: realized trading volume, and the disclosed settlement-source hierarchy. If volume is thin and settlement rules stay vague, this is a narrative, not a market. Chop is for positioning. This is a position worth watching — not taking. Yet.