Prediction Markets Go Institutional: Bots Meet Wall Street

On the first trading day of the fourth quarter, the biggest venue in prediction markets is not raising money to offer better odds. It is raising money to look more like a bank. Kalshi, the CFTC-regulated exchange that spent 2025 as a sports-betting curiosity, is in talks to raise roughly $1 billion at a $40 billion valuation, about 82% above the $22 billion it carried five months earlier, with Sequoia Capital and Wellington Management lined up to lead the round, according to reporting published on 30 September. Polymarket, its on-chain rival, is separately working on a round near $21 billion, per DeFi Rate.

The strange part, for a market the public still files under election bets and viral long shots, is who got there first. The order flow on these venues is dominated less by retail gamblers or Wall Street desks than by software. Autonomous AI agents place a large share of the trades, quote the tightest books, and never log off. For most of the last two years that was the whole story: machines quietly annexed a crypto backwater. The story in late 2026 is different. The backwater is being rezoned as financial infrastructure, and the firms that build financial infrastructure (exchanges, prime brokers, proprietary trading shops, index providers) are moving in on top of the bots. That is the institutional turn, and it is rewriting what it means to be a prediction-market agent.

From a crypto sideshow to financial infrastructure

Rewind eighteen months and prediction markets were a rounding error. Monthly volume across the sector was around $1.2 billion in early 2025; by January 2026 it had cleared $20 billion, according to an analysis from blockchain-intelligence firm TRM Labs, which also clocked more than 840,000 unique wallets trading in a single month and a single-day record above $425 million during the February 2026 surge in Iran-related markets. The mix shifted along the way, from crypto wagers toward geopolitics, US politics, and sports, the categories that pull in mainstream attention and mainstream money. An event contract is a simple instrument: it pays $1 if a stated outcome happens and $0 if it does not, so its price reads directly as a probability. That simplicity is exactly what makes it attractive to institutions, which can treat a cleanly defined binary as a hedge, a sentiment feed, or a tradeable view on the world.

The pivot has a slogan in the financial press. As CNBC put it in June, the individual traders who drove Kalshi’s rise have given way to a push for Wall Street. The ambition is openly enormous. Bernstein analysts led by Gautam Chhugani project the sector growing from roughly $51 billion in 2025 to a pace near $240 billion in 2026 and about $1 trillion by 2030, a compound growth rate close to 80%, per CNBC. Kalshi co-founder and chief executive Tarek Mansour has pitched event contracts in the same register, describing them as something that could become a trillion-dollar market as the current round came together. Whether or not those figures land, the behavior they are driving is unmistakable: the venues are rebuilding themselves as exchanges, not casinos, and recruiting the plumbing to match.

The money: a $40 billion race

The clearest sign of the institutional turn is the cap table. Kalshi’s jump to a $40 billion target, with Sequoia and Wellington leading and Tiger Global and Dragoneer named as possible newcomers, would value a six-year-old event-contract exchange at more than most regional banks. The round is being justified by growth that reads like a derivatives startup rather than a betting site: Kalshi’s annualized trading volume tripled from about $52 billion to $178 billion over six months, institutional trading activity climbed roughly 800%, and the platform now accounts for more than 90% of US prediction-market activity, per the same 30 September report. Mansour has said the company is weighing an eventual public listing, with Sequoia partner Alfred Lin on its board.

What the investors are really paying for is a moat built from flow and data rather than from the house edge. A venue that clears the most volume attracts the most liquidity, which tightens its prices, which pulls in still more volume, including the professional and algorithmic flow that now dominates. Add exclusive data-distribution rights and brokerage partnerships on top, and the result looks less like a betting site and more like an exchange franchise, the kind of asset that has historically commanded rich multiples. That is the bet behind the billion-dollar valuations.

Polymarket’s path runs through crypto rather than a US exchange licence, but the trajectory rhymes: a round near $21 billion, up from $15 billion in the spring, with ICE already on the register. Behind both sits a structural shift that would have been unthinkable in 2024, when prediction markets were synonymous with offshore betting. The table below lays out where the two leaders stand as the quarter turns.

Why the machines got there first

Before any prime broker showed interest, autonomous agents had already made prediction markets their native habitat. On Polymarket, AI agents account for more than 30% of active wallets, and 14 of the top 20 wallets by activity are bots, according to CoinDesk. The flagship example, Polystrat, was launched in February 2026 by Valory, the company behind the Olas agent network; in its first month it placed more than 4,200 trades and returned as much as 376% on a single position. Agents are also more likely to finish ahead: roughly 37% of Polystrat’s agents ran a positive profit and loss, against the 7% to 13% of human traders who manage the same.

David Minarsch, Valory’s chief executive, describes Polystrat as “an autonomous AI agent that trades on Polymarket 24/7 on behalf of its human user.” He is candid about the limits: simply prompting an off-the-shelf model with a market, he told CoinDesk, “usually results in outcomes no better than a coin-flip,” while a purpose-built workflow can push predictive accuracy past 70%. The deeper reason agents thrive here is structural. A prediction market has a bounded action space (buy yes, buy no, size the bet), an objective and scheduled payout, and a public order book that runs around the clock, which is close to an ideal environment for software. The economics reward specialization, too: on the Olas network, a trading agent can buy probability estimates from other agents in a marketplace, one of the few working examples of the machine-to-machine commerce the rest of crypto keeps promising. TRM Labs’ segmentation shows the signature of that machine presence: high-frequency market makers placing more than 10,000 trades each accounted for about 35% of all trades despite being a sliver of accounts. The agents are not a novelty on these venues. They are the liquidity, and that is precisely why institutions want in.

The perps pivot: a prediction market becomes a derivatives exchange

If one product captures the institutional turn, it is the perpetual future. In May 2026 the CFTC issued its first approval for a crypto perpetual future at a regulated US venue, clearing KalshiEX to list a Bitcoin perpetual, per CoinDesk. Perps are the dominant instrument of offshore crypto trading, and bringing them onshore under a federal regulator is a serious escalation from binary event contracts. The appetite was immediate: Kalshi’s Bitcoin perpetual crossed $1 billion in volume within a week of launch and topped $5.5 billion inside two weeks, according to CNBC, and the contracts have cleared tens of billions in notional since.

From there the menu expanded quickly: gold and silver perpetuals after further CFTC approval in September, with filings in for a major US stock index and for copper. In other words, a company that started by letting people bet on elections now runs a growing, federally regulated derivatives exchange. For traders used to decentralized venues, the shape is familiar; our field guide to on-chain perpetual futures covers the mechanics, and the leverage perps introduce is the same leverage that can erase a position, as our explainer on what happens when you get liquidated lays out. The table traces Kalshi’s march from event contracts into derivatives.

Perps also happen to suit autonomous agents better than one-off event contracts do. A perpetual never expires, so it trades continuously, which rewards the one participant that is always online; its funding rate is a clean, machine-readable signal of where leverage is leaning; and its deep, fast order book is exactly the kind of venue a market-making bot is built to work. The instrument that pulls in Wall Street desks, in other words, also hands the agents a bigger and more liquid playground. That is the pattern of the whole institutional turn: every feature added for professionals is a feature the machines can use first.

ICE and the data play: the betting is the byproduct

The most telling institutional bet on prediction markets is not on the betting at all. Intercontinental Exchange, the owner of the New York Stock Exchange, has committed up to $2 billion to Polymarket across an initial 2025 stake and a $600 million follow-on, and its stated thesis is data, not wagering. ICE became the exclusive global distributor of Polymarket’s event-driven data to institutional clients, and in February 2026 it launched a Polymarket Signals and Sentiment tool that pipes real-time prediction-market probabilities through the same ICE Consolidated Feed that already carries securities prices, fundamentals, and corporate actions. In effect, the deal turns Polymarket’s odds into a licensed data product, the same status ICE already grants to equity and commodity prices.

ICE founder and chief executive Jeffrey Sprecher has told analysts that the appeal is the information the crowd generates, which the company can normalize against its reference databases and sell to clients alongside its existing market data, rather than any expectation about the betting multiple itself. It is a revealing framing. To one of the largest exchange operators on the planet, a prediction market is less a place to gamble than a continuously updating sensor for real-world probabilities, and the AI agents trading on it are the mechanism that keeps that sensor calibrated. When the agents price a market efficiently, ICE’s data product gets sharper. The bots and the data business are, in that sense, the same trade, which is why the institution that could have ignored prediction markets instead paid up for the feed.

The pipes: Robinhood, Alpaca, and institutional distribution

Infrastructure is nothing without distribution, and 2026 was the year the pipes got built. Kalshi’s partnership with Robinhood put event contracts in front of a brokerage with tens of millions of funded accounts, through an in-app hub that grew from election and March Madness markets into a full sports lineup; at the peak, Robinhood came to account for more than half of Kalshi’s total trading volume. In August the exchange went further, partnering with brokerage-infrastructure firm Alpaca to route its CFTC-regulated event contracts to roughly 300 financial institutions and 14 million brokerage accounts worldwide, according to CNBC. Alpaca had registered a derivatives arm with the CFTC weeks earlier, specifically so it could carry event contracts through the same rails brokers already use for stocks, options, and crypto.

The institutional milestones kept coming. Kalshi completed what was billed as the first block trade on a prediction market, matching a Texas hedge fund with a market maker on California carbon-allowance contracts, as CNBC reported. It also shipped Kalshi Pro, a desktop terminal with a screener across roughly 2,000 markets, charting, and a perps interface, aimed squarely at proprietary trading shops rather than casual bettors. Each of these is a brick in the same wall: a prediction market that wants to be traded like an asset class, by the same desks and through the same brokers that trade everything else. For the autonomous agents that built the early liquidity, it means the neighborhood is filling up fast.

Who trades now: retail, desks, and machines

The institutional turn does not evict the agents; it crowds the book. For most of the easy-money era, a competent bot on Polymarket faced slow, emotional, part-time human counterparties, and the edge was there for the taking. As prop desks, quant shops, and brokerage flow arrive, that edge compresses. The counterparty on the other side of an agent’s trade is increasingly another professional, often another agent, and in some cases a human-run desk that has itself deployed AI to copy or generate signals. The result is a three-sided market: retail providing noise and liquidity at the fat head of popular events, institutional desks bringing capital and sharper pricing, and autonomous agents working the long tail and the clock.

The uncomfortable implication is that agents now face the same adverse selection they used to impose on retail. A naive model that could print money against distracted humans in early 2026 gets picked off by faster, better-capitalized opponents today. That is why the winning agents are consolidating into professional operations, and why proprietary trading firms have started staffing dedicated prediction-market desks. Prime brokers and quant shops have built the access to match, and some desks now run their own AI agents to copy or originate signals, so the machines increasingly trade against other machines rather than against the distracted retail flow that made the early returns so easy. The firms tend to describe the edge in modest terms, confidence about the direction of a market rather than its magnitude, but spread across thousands of constantly repricing contracts, that is enough to build a business on. The table sketches the three classes of participant now sharing the order book.

The volume question: how real is the growth?

Institutionalization raises the stakes on a question that was easy to wave away when prediction markets were tiny: how much of the volume is real? The issue came to a head in the final days of September. A market observer flagged on X that Kalshi’s Ether perpetual had posted about $539 million in 24-hour volume against only $3.1 million in open interest, a ratio that is hard to explain through ordinary trading. A CNBC review found that on 20 September nearly half the dollar volume on the Ether perpetual came from trades sized between roughly $5,495 and $5,505, and The Wall Street Journal reported the CFTC was examining the pattern, with more than $5 billion in such trades over a single month, per The Block.

Kalshi denies any wash trading, saying the repeated prints came from market makers posting fixed quotes that faster traders hit, and that it is not under investigation. It is nonetheless winding down its Volume Incentive Program no earlier than 13 October, a scheme that paid traders for activity and that critics argue can manufacture the appearance of it. The company still posted a record $52.98 billion in volume in September. The episode matters beyond one contract. Independent microstructure research on Polymarket has found that traders acting as their own counterparty is usually a small share of activity but spikes sharply in some markets, and the same leverage that juiced perps volume is the leverage that forces liquidations when it unwinds. The deeper worry is reflexive: incentive programs and leverage can both inflate the very volume figures that justify the next funding round, so the growth story and the volume questions end up feeding on each other. For institutions pricing these venues as data and as an asset class, the integrity of the tape is not a footnote. It is the product they are buying.

A market institutionalizing without a rulebook

The institutional money is arriving before the law has settled what it is buying. US prediction markets live under the Commodity Futures Trading Commission, which regulates event contracts as a species of derivative; the Securities and Exchange Commission only touches the crypto tokens around the edges, such as the governance tokens of the oracle and agent networks, not the wagers themselves. That CFTC framing is now contested in the courts. Appeals panels have split three ways on whether sports event contracts are federally regulated swaps or state-regulated bets: the Third Circuit sided with Kalshi, while the Ninth Circuit and, on 25 September, the Sixth Circuit sided with the states, as Sportico detailed. Several state attorneys general have separately sued to shut the venues down as unlicensed gambling, while the CFTC insists it holds exclusive federal authority over event contracts, a standoff that pits Washington against the states just as the institutional money arrives.

That split is a direct invitation to the Supreme Court. New Jersey, Robinhood, and Crypto.com have all petitioned for review, while Kalshi seeks a rehearing before the full Ninth Circuit; a cert grant could put the question on the docket for the term running into mid-2027. Underneath it sits the CFTC’s own proposed rewrite of Rule 40.11, which would set a case-by-case test for when an event contract touches a prohibited activity such as gaming, a rule still in draft under Chair Michael Selig. For how this fits the wider policy calendar, see our Q3 regulatory scorecard and Q4 countdown. The irony is hard to miss: capital is pricing these venues at tens of billions of dollars while the most basic legal question, whether they are exchanges or sportsbooks, remains open.

Event-contract ETFs and the productization frontier

The next wrapper is already in view. As volumes climbed, regulators began scrutinizing proposals to package prediction-market exposure into exchange-traded products, the same financialization path that took crypto from self-custody wallets into ordinary brokerage accounts. The logic is identical to the one that drove the spot-crypto ETF wave, and the same mechanical questions apply once a product trades on an exchange; our look at how ETF approvals reshape the volatility surface explains why the wrapper changes the trade underneath it. Whether or not event-contract ETFs clear in the near term, their mere consideration signals how far the asset class has traveled: from a Discord curiosity to something a compliance desk can file a prospectus on.

The productization also closes a loop back to the agents. An exchange-traded product or a structured note needs a reliable, continuously priced underlying, and that underlying is exactly what a book full of 24/7 trading bots provides. The more the venues are wrapped and resold, the more they depend on the machine liquidity underneath, which means the institutions buying in are, knowingly or not, buying a market whose prices are set mostly by software.

When agents carry institutional size: the trust problem

Scaling the money scales the risk. An agent that loses a hobbyist’s $500 is a bad afternoon; an agent wired into a desk running institutional size is a different exposure, and its failure modes are specific to autonomy. Agents that hold their own keys can be drained through prompt injection, tricked into signing a malicious transaction, or compromised at the device layer, the same hazards that plague on-chain agents everywhere. The question of whether you can safely hand a wallet to a language model is unresolved, as we examined in our look at trusting an Eliza agent with a wallet.

There is also a market-structure risk that grows with institutional adoption: correlation. When many desks run similar models against the same public order book and the same data feeds, their agents tend to move together, amplifying swings and inviting the order-flow games that already crowd busy markets. A monoculture of near-identical bots is efficient until it is suddenly, simultaneously wrong. Institutions bring risk limits and capital buffers that a lone hobbyist lacks, which helps; they also bring size, which makes any correlated failure bigger. The trust problem does not shrink as prediction markets grow up. It changes shape.

Tokens versus platforms: the value-capture gap

For crypto investors, the institutional turn exposes a blunt divide between the platforms and the tokens. The equity value is concentrating in private companies: Kalshi near $40 billion, Polymarket near $21 billion. The public tokens attached to the ecosystem tell a different story. UMA, the token behind the optimistic oracle that resolves most Polymarket outcomes, trades around $0.41 for a market capitalization near $38 million, and OLAS, the token of the Olas agent network behind Polystrat, sits near $0.048 for about $14 million, per CoinGecko and its listing for Autonolas. Both are micro-caps a rounding error away from the valuations their host platforms command.

The lesson is one the AI-agent sector has learned the hard way: framework quality does not equal token performance. Olas runs real infrastructure, an agent-to-agent marketplace, millions of on-chain transactions, and the Polystrat agent that beats most humans, yet its token is still a micro-cap. UMA arbitrates markets worth far more than its own market value. The institutions pouring into prediction markets are buying equity, data rights, and order flow, not governance tokens, and that is where the value is accruing. Anyone conflating the two is reading the wrong scoreboard. For the token holder, the agent networks and oracles are essential plumbing with weak value capture; for the venture investor, the exchange that routes the flow is the prize. The institutional turn has made that gap wider, not narrower.

What to watch in Q4 2026

A handful of questions will decide whether the institutional turn is a durable re-rating or a late-cycle top. The ones worth tracking through the quarter:

• Whether Kalshi’s $40 billion round actually closes, on what terms, and how quickly Polymarket’s follows.
• A Supreme Court cert decision on the swap-versus-bet question, which would set the ground rules for every institutional participant.
• The final shape of the CFTC’s Rule 40.11, which will define which markets are even allowed to exist.
• How the wash-trading scrutiny resolves once the Volume Incentive Program ends in October and reported volumes have to stand on their own.
• Whether autonomous agents keep their edge as prop desks and brokers scale up, or get absorbed into human-run operations as execution tools.

The bottom line

Prediction markets grew up in 2026. The bots built the liquidity, and now Wall Street is buying the house. What was a crypto curiosity two years ago is being priced, plumbed, and productized like a real asset class, complete with perpetual futures, data feeds sold through exchange terminals, brokerage distribution, and nine-figure funding rounds. The AI agents that dominated the early game are not going away; they are becoming one layer in a deeper stack, sharing the book with the same institutions that run every other market. The open questions are the ones that always follow fast money: whether the volume is real, whether the law will bless it, and whether a market run largely by correlated machines is as robust as its new valuations assume. The answers will arrive over the next few quarters. The bots will keep quoting through all of it, because that is what they do. For now, the most automated corner of crypto has become its most institutional, and that paradox is the whole story.

Frequently Asked Questions

Do AI agents really dominate prediction markets?

Yes. On Polymarket, autonomous AI agents account for more than 30% of active wallets, and 14 of the top 20 wallets by activity are bots, according to CoinDesk. Agents also tend to be more profitable than humans, and high-frequency market-making bots place a large share of all trades, which is why institutions treat them as core liquidity rather than a novelty.

What is behind Kalshi’s $40 billion valuation?

Kalshi is in talks to raise about $1 billion at a $40 billion valuation, up roughly 82% from $22 billion five months earlier. The round is driven by surging volume, with annualized trading tripling to about $178 billion over six months, and by an 800% jump in institutional activity as the company pushes from retail event contracts into regulated derivatives.

Are prediction markets gambling or trading?

Legally, it is unsettled. US prediction markets operate as event-contract exchanges under the CFTC, which treats the contracts as derivatives, but appeals courts have split three ways on whether sports contracts are federally regulated swaps or state-regulated bets. The question is widely expected to reach the Supreme Court, with cert petitions already filed by New Jersey, Robinhood, and Crypto.com.

What are Kalshi perpetual futures?

They are perpetual futures contracts, the dominant instrument of offshore crypto trading, now offered onshore under CFTC approval. Kalshi listed the first CFTC-approved crypto perpetual, a Bitcoin contract, in 2026, then added gold and silver perpetuals and filed for a stock index and copper, turning a prediction market into a regulated derivatives exchange.

Is Polymarket and Kalshi trading volume real?

Some of it is being questioned. In September 2026, analysts and a CNBC review flagged unusual patterns on Kalshi’s Ether perpetual, and the CFTC was reported to be examining repeated trades that may have inflated volume. Kalshi denies any wash trading and is winding down a volume-incentive program, but the scrutiny shows why tape integrity matters as the market institutionalizes.

By Marcus Okafor, senior markets writer at HOGE Wire, covering AI agents, prediction markets, and market structure.