AINEWS
Meta tried to buy Kalshi. Now it's building a fake-money version, run by AI.
Prediction markets are sites where people wager real money on real world events, from elections to sports games.
Prediction markets are sites where people wager real money on real world events, from elections to sports games.
Before Mark Zuckerberg directed Meta engineers to build a standalone prediction market app, he tried a simpler route: buy the leader. In a 2025 meeting with Kalshi chief executive Tarek Mansour, Zuckerberg proposed acquiring the company, according to three sources familiar with the talks (NPR via WDIY). The takeover never advanced, and the failure to land Kalshi now looks like the opening act of a familiar Meta playbook: when the company cannot own the category, it tries to clone it instead.
Prediction markets are websites where users wager real money on the outcomes of real-world events, from elections to football games. Kalshi is the largest U.S.-regulated platform of this kind, and Polymarket is its main crypto-based rival. Both let traders express a view on the probability of something happening, with prices reflecting collective belief. Meta's in-house answer, internally known as Arena, keeps the betting interface but discards the two ingredients that make the model work. There is no real money on the line, only a daily allotment of virtual currency, and the company's own artificial intelligence both generates the questions and decides who wins (NPR).
That structural choice is the news. A prediction market's value comes from the price signal: traders risk real capital and an independent resolver settles the outcome. Strip both out, and the product becomes something closer to a quiz game curated by a corporation, with the corporation's AI acting as bookmaker, judge, and jury. The original directive came from Zuckerberg himself, who told employees to create a prediction markets app, according to internal Meta documents reviewed by The New York Times (NYT; Reuters). Forbes also reported Zuckerberg's interest in a prediction market app (Forbes).
Why the Kalshi talks collapsed is now a contested story. One narrative is that Mansour refused to sell. Another is that Meta walked away, judging Kalshi's legal and ethical exposure too messy to inherit. NPR attributed both versions to sources familiar with the discussions; neither side has confirmed publicly. The disagreement matters less than the result: Meta is now in the position of building a competitor to a company whose valuation has roughly tripled in six months. Kalshi raised a Series E at an $11 billion valuation in December 2025 (Kalshi newsroom), with subsequent rounds lifting it to roughly $22 billion by May 2026 and talks reportedly pushing toward $40 billion (CryptoBriefing).
Columbia Law professor Tim Wu, a former tech adviser to President Biden, frames Arena as the latest entry in a longer list. "Meta seems to clutch at every shiny object, fail, and move on to the next shiny object, without consequence," Wu told NPR, citing the company's pullback from the metaverse and the abandoned Libra cryptocurrency project. Asked specifically about a fake-money casino, Wu added, "I can't imagine a casino app with fake money is going to be much of a thrill" (NPR via WDIY). The critique doubles as a read on Arena: if the thrill is the point, removing the stakes removes the product.
The political reception has already begun. Senator Richard Blumenthal criticized Meta's plans publicly, drawing a line from Instagram's engagement-optimizing features to the introduction of prediction markets, a sign that even a play-money version is going to draw regulatory attention in a category already under CFTC scrutiny. TechBrew reported that Arena's launch date, monetization path, and final feature set are not yet public (TechBrew).
The framing is internally consistent: when Meta cannot own the leader, it ships a version that costs less and ships faster, then asks the user to call it the same product. What an AI-curated, no-real-money version of a prediction market actually is, is the open question. If Arena trains users to treat AI-generated questions and AI-declared winners as a market, it changes what the category means before any regulated competitor gets the chance to define it. If users see through the game, Meta will have shipped another Libra: a category toy with the category's name on it.