Prediction Markets Eat DeFi
Political analysts, oddsmakers, quants, and bots walked onto the same order book
Prediction markets are being compared to options, perpetual futures, sportsbooks, exchanges, news outlets, polling firms, hedge fund research desks, insurance products, sentiment indicators, and intelligence briefings.
Financial markets have always separated their participants by infrastructure. Futures pits have hedgers and speculators but they share a theory of what a contract is and how it settles. Sportsbooks have sharps and squares but they both understand the vig and the line. DeFi protocols have LPs and arbitrageurs but they share the language of pools and slippage. Each developed its own ecology, its own definition of edge, and none of them ever had to share an order book with anyone from outside their tradition.
Over the past year prediction markets absorbed enough infrastructure from each of these that participants from all of them can now show up to the same order book and see something familiar enough to trade. Each one looked like a routine technical upgrade at the time, but each one let a different population walk in.
How they all ended up on the same book
Kalshi tokenized its entire order book onto Solana in December. Every event contract is now mintable as an SPL token, settling in USDC. A quant looks at that and sees a simplified option. A bookmaker sees a no-vig line. A DeFi protocol sees a composable primitive. Same contract, three different products. Then Builder Codes let any developer embed these markets into their own app and earn fees on the flow. Drift BET started integrating prediction mechanics into its perps platform. edgeX expanded from perps into prediction markets. The boundary is dissolving from both directions, not because anyone planned it but because the infrastructure made it easy enough to walk through.
What you end up with is an order book where nobody shares a theory of what the market is for and nobody can tell who is on the other side of any given trade.
A geopolitical contract like US strikes on Iran, currently over three hundred million in total volume on Polymarket, is useful to see what this collision looks like in practice.
The strikes market has over thirty million in recent volume. The deal market on the same situation has seven hundred thousand. That volume gap probably tells you more about who is on each book than about the likelihood of either outcome. The political analyst on this book has spent years modeling how administrations escalate. She reads OSINT flight data, tracks diplomatic language, prices the contract as a probability estimate of a real-world event. The quant desk sees the same contract and sees structural alpha. The Becker dataset, seventy-two million trades now public under MIT license, shows takers losing systematically across almost every price level. Spreads are wide, flow is unsophisticated relative to rates or equity options. DeFi agent is not predicting anything about Iran. It is arbitraging price discrepancies between the on-chain SPL token and the centralized Kalshi order book, routing flow through Builder Codes for rebate optimization. The contract could be about anything. And the journalist covering the Iran story is embedding the market’s live probability into her reporting the way she used to embed a poll or an analyst quote.
Traditional markets solved this problem a long time ago. Prediction markets have not even started. On a derivatives desk you can infer flow type from order size, timing, and venue. On a sportsbook an experienced bookmaker reads whether she is facing a sharp or a square from the pattern of the action. On a prediction market order book you might be trading against a quant running models calibrated on seventy million historical trades, or someone who scraped an unreleased result from a Wordpress preview page, or an AI agent executing through an API you have never heard of, or a fan who just wants skin in the game. There is no way to tell which one. The counterparty problem is not theoretical.
A trader on a recent Spotify market learned this when he built a data model, sized up a confident position, and lost because someone on the other side had found the answer on a preview page before the official release. The counterparty was not smarter. They were earlier. He had no way to know that until the money was gone.
Nobody knows who they're trading against
The fifteen-minute Bitcoin contracts Kalshi launched in January show how the absorption works mechanically. Same contract settling on CF Benchmarks Real-Time Indices. A crypto trader uses it as an option without the Greeks. A sportsbook user treats it as a peer-to-peer bet without the house edge. A Robinhood retail investor gets derivatives exposure without margin requirements. Three different products that happen to be one contract on one set of rails.
NPR published a story this week about a participant type that does not fit any traditional category. Traders buying television antennas for fractional-second latency edge during live events. A teacher who made ten thousand dollars from a seven hundred dollar Billboard bet using unreleased streaming data.
The edge in prediction markets turns out to come less from processing public information and more from presence, from being physically or digitally somewhere the information exists before the rest of the market knows it does. Which probably explains why AI models, despite having access to all public information and the ability to process it faster than anyone, have not dominated prediction markets. The incumbents can see the volume moving but they cannot see what is driving it.
FanDuel's CEO went on television to explain why she’s not worried
Sportico reported that the emerging battle on prediction exchanges is between Wall Street trading desks and veteran sportsbook oddsmakers, two groups who have never competed for the same order flow.
When FanDuel’s CEO went on CNBC the same week to call prediction markets “closer to daily fantasy,” she was doing what incumbents do when they cannot classify the threat: pulling it back into her tradition’s vocabulary, because the alternative is admitting her counterparties now include quant desks and DeFi protocols and she has no framework for competing with them.
When analyst questions on Robinhood’s Q4 earnings call kept coming back to event contracts and three hundred million in annual revenue, those analysts were not asking what is this. They were asking how much of our existing flow migrates here. That is a different conversation, and it tells you the infrastructure underneath has become familiar enough that a brokerage’s analysts are no longer trying to understand event contracts but trying to figure out how much of their existing business ends up there. And the range of questions that can pull these populations onto the same book keeps growing.
Polymarket just made liquidity rewards permissionless. Anyone can now attach incentives to any market to subsidize tighter spreads. Permissionless market creation and creator fees are next. Before this, the platform decided which questions got liquid enough to trade. Now anyone can pay to make a market tradeable, and each new market category drags a different mismatch onto the book. Geopolitical markets resolve through human-judged oracles that quants have no model for. Housing markets settle against on-chain Parcl indices that give oddsmakers nothing to work with. You are not just paying for a forecast anymore. You are paying traders to go find things out.
The Super Bowl processed one point three four billion across prediction platforms in a single day. Total volume quadrupled year over year. Kalshi is valued at eleven billion, Polymarket at nine after ICE put in two billion. Substack launched native Polymarket embeds this week. When a newsletter platform builds native integration for prediction market contracts, you are no longer watching a financial product grow. You are watching infrastructure absorb another medium.
The CFTC chairman says the commission “will no longer sit idly by while overzealous state governments undermine the agency’s exclusive jurisdiction.” Massachusetts won an injunction against Kalshi. Nevada got a restraining order against Crypto.com.
The regulatory fight is real but it follows from something more basic. When participants from five different traditions are trading the same contracts, the question of which regulatory framework applies does not have a clean answer, because the traditions those frameworks were built for never shared an order book before.
There are live markets on military strikes, fifteen-minute crypto movements, AI rankings, the World Cup, on-chain housing indices, and whether a character gets engaged on Bridgerton. Five years ago these would have been on five different platforms governed by five different frameworks traded by five populations who would never have encountered each other.
Now they share an order book, and the order book does not care that they disagree about what it is. It just needs enough of them to show up on each side.