Meta Plans an AI-Powered Prediction Market App to Rival Kalshi and Polymarket

Meta Plans an AI-Powered Prediction Market App to Rival Kalshi and Polymarket

The Vault — AI Edition | Business | June 25, 2026

Mark Zuckerberg wants a piece of the fastest-growing corner of finance, and he is betting that a large language model can run it. According to internal company documents reported this week, Meta has assembled a team to build a standalone prediction-market app — codenamed "Antwerp" and "FBForecast" internally, and referred to in some reporting as "Arena" — that would let users wager on the outcomes of real-world events the way they already do on Kalshi and Polymarket. The twist that sets Meta's effort apart: its Llama large language model would automatically generate the markets, recommend them to users, and resolve which side won, in near real time and with no human in the loop.

It is an audacious move into a sector that has gone from fringe curiosity to one of Wall Street's hottest stories in under two years — and one that immediately rattled the incumbents it threatens.

What Meta Is Building

The app, according to the documents, would be walled off from Meta's core properties — Facebook, Instagram, WhatsApp, and Messenger — and operate as its own product. Llama would scan trending topics across elections, sports, geopolitics, and viral internet moments, then spin up binary yes-or-no questions for users to predict. The same model would make "personalized market recommendations" to each user and, critically, deliver the final ruling on whether a market resolved yes or no.

At least at launch, users would not stake real money. Instead, Meta plans a video-game-style points system to reward accurate forecasters, an approach that keeps the product clear of the thorniest gambling-regulation questions. But the documents reportedly leave the door open to real-money betting later, which would put Meta squarely in the regulated derivatives arena occupied by Kalshi.

The design is also a notable echo of Meta's own history. In 2020, the company launched a crowdsourced prediction app called Forecast, where users guessed at events including the trajectory of the COVID-19 pandemic. Meta wound it down two years later. This time, the differentiator is generative AI doing the work that human moderators and community resolution committees handle elsewhere.

A Sector on Fire

The timing is no accident. Prediction markets have exploded: the sector grew from roughly $50 billion in total trades during 2025 to more than $130 billion in just the first half of 2026. Kalshi, the U.S. regulated leader, generated $21.1 billion in volume in June 2026 alone, against Polymarket's $9.7 billion.

Kalshi's valuation tells the story even more starkly. The company was worth about $5 billion in early 2025, $11 billion by December, $22 billion in May 2026, and is now in talks to raise at roughly $40 billion — an eightfold jump in about 18 months. Bulls think the ceiling is far higher. Kalshi chief executive Tarek Mansour has framed the opportunity in sweeping terms: "Event contracts could become a trillion-dollar market, and we're still in the early stages of that transition." Coatue's Philippe Laffont has echoed the trillion-dollar framing, suggesting today's rich valuations may still understate the prize.

A trillion-dollar market that runs on viral attention and real-time information is, almost by definition, a market Meta believes it should own.

Why It Matters

Meta's entry is the clearest sign yet that the AI giants see prediction markets not just as a financial product but as an AI product. A prediction market is fundamentally a question-and-resolution engine — exactly the kind of task generative models are being pushed to automate. By letting Llama write the questions and call the outcomes, Meta is proposing to compress a workflow that today depends on human oversight and dispute windows into an automated pipeline running at the speed of the feed.

That is also where the risk concentrates. In the reported design, no human reviews Llama's calls and no challenge window exists to dispute them — handing an AI model the role of arbiter of truth on contested events like elections and geopolitics. Trust is the obvious vulnerability. A Forrester survey published June 25 found that 56% of respondents do not trust Meta with a prediction-market product, and roughly a third said the company's involvement makes them less likely to try the app than they would be with any other operator.

Investors registered the threat regardless. On the initial reports, shares of DraftKings fell more than 2% and Flutter Entertainment, the parent of FanDuel, slid nearly 2% before recovering, with Robinhood also caught in the move. Analysts were more measured: a Jefferies analyst characterized Meta's entry as a manageable competitive threat for the incumbents rather than an existential one — a reminder that distribution muscle does not guarantee a winning product, especially from a company carrying Meta's trust baggage.

What to Watch Next

The open questions now are timing, money, and regulation. Meta has not publicly confirmed a launch date, and a points-only version sidesteps the regulatory gauntlet that Kalshi spent years clearing with the Commodity Futures Trading Commission. The moment Meta turns on real-money wagering — if it does — the calculus changes entirely, dragging in the CFTC, state gaming regulators, and renewed scrutiny of an AI deciding what is true. Watch, too, for how Llama's resolutions hold up under pressure: a single high-profile blown call on a disputed election or sports outcome could validate every doubt the Forrester numbers already surfaced. Meta has the reach to make a prediction market enormous overnight. Whether anyone trusts the machine calling the shots is the bet that actually matters.

---

Sources: NPR; Gizmodo; CNBC; Forrester; CoinDesk; Cryptopolitan; Isla Public.