Meta's Push into Prediction Markets Is a Data Play
· By Tyler Jacobsma
The points-not-cash design looks timid. It's the most valuable part, because the prize was never the wager. It's the belief, and what Meta's ads and AI businesses can do with it.
There is an easy way to read the news that Mark Zuckerberg has a team building a prediction-market app, internally called Arena, that won't let anyone bet real money. The easy read is that Meta is being cautious — cloning Polymarket and Kalshi the way it cloned Snapchat and TikTok before them, but flinching at the regulated, gambling-adjacent part and shipping a watered-down points game instead. Forecast, Meta's last attempt at exactly this, launched in 2020 and was dead by 2022. So you can be forgiven for filing Arena under "Zuckerberg chases a trend, again."
I'd file it somewhere else. The decision to launch without real money isn't the compromise in this plan. It's the plan. And to see why, you have to stop thinking about Arena as a betting product and start thinking about it as a data-collection instrument pointed at the two businesses Meta actually cares about: advertising and artificial intelligence.
Consider what Meta currently knows about you. It knows what you clicked, what you lingered on, what you liked, who you follow. That's an enormous amount of behavioral exhaust, and Meta has built the most valuable ad business in history on top of it. But notice what's missing from that list. A like tells Meta you reacted to something. A comment tells Meta you hold an opinion. Neither tells Meta whether you were right, or how confident you were when you weren't.
A prediction does. A forecast is the first native social action that yields a scored, falsifiable, calibrated measurement of what a person actually believes will happen. When a user puts 70% on a team winning, or a rate cut landing, or a film clearing some box-office number, that's not a vibe, it's a quantity, and the world eventually grades it. Multiply that by the audience Meta can funnel in from Facebook and Instagram, and you get something no social platform has ever held: a continuously resolving map of what billions of people expect the future to look like, sorted by how often each of them turns out to be correct.
For the ads business, that is a different and richer class of signal than anything in the current stack. Advertisers don't really pay for engagement; they pay for intent, and intent is fundamentally a belief about the future, that you'll need a flight, that the truck is worth buying, that the election outcome changes what you'll spend on. A platform that can see which users genuinely expect a recession, or a team to win, or a product category to take off, is sitting on forward-looking intent data rather than the backward-looking behavioral kind everyone else is forced to infer from clicks. Knowing who is reliably well-calibrated is itself a targeting and trust primitive Meta doesn't have today.
Here's the part that makes the timing more than coincidental. Meta is in the middle of the most expensive AI buildout of any company on earth, and the binding constraint in frontier AI right now isn't compute or talent — it's high-quality, hard-to-fake training data, especially the kind that teaches a model to reason about uncertain future events. Resolved human forecasts are almost a perfect fit for that gap. They come pre-labeled, because reality supplies the answer key, and they arrive in exactly the "predict the outcome from messy real-world context" format that reasoning models are starved for. Arena, in that light, is a perpetual self-resolving labeling machine in which users volunteer the labels for status points instead of being paid for them. Meta gets to train on humanity's forecasting record and bill the data-collection cost to a free game.
This is also why real money would make the product worse for Meta's purposes, not better. Cash drags in the Commodity Futures Trading Commission, know-your-customer requirements, and the awkward fact that most Polymarket users lose money over time , a selection effect that narrows and distorts the participant pool. Points keep the funnel wide, the regulatory surface small, and the data clean. A points-only Arena can launch quickly and sidestep CFTC jurisdiction entirely, which is not a hedge so much as a feature. The standalone-app structure, fed by Instagram but walled off from the money, is precisely how you'd build a data surface you intend to wire into the ad and AI stack later, rather than how you'd build a destination gambling brand.
The incumbents should worry about the geometry of this, even though Meta isn't attacking them head-on. The threat is lateral. A free points game that absorbs the casual attention — the very users whose participation makes prediction-market prices liquid and accurate — could quietly drain the activity that Polymarket and Kalshi depend on, while Meta walks off with the byproduct it actually wanted and leaves the regulated trading business behind. The market reaction was telling enough: DraftKings fell more than 2% on the report and is down 37% this year, with 2026 revenue guidance of roughly $6.5 billion to $7 billion against the $7.33 billion analysts had penciled in. The category was already proving harder to monetize than the incumbents promised. Meta's arrival doesn't help.
Now, the honest objection, because it's a serious one: skin in the game is what makes prediction markets work. The reason Kalshi and Polymarket prices mean anything is that people stand to lose money for being wrong, and that cost forces honesty into the number. Strip out the money and you may get idle guessing, exactly the fate that befell Forecast. If Arena's forecasts carry no penalty for being wrong, the calibration data is worth considerably less than the bullish version of this argument assumes.
The reply is that Meta and Kalshi are optimizing for different things. Kalshi needs prices to be correct, which requires skin in the game. Meta needs beliefs to be profiled at scale, which doesn't. For price discovery, a thousand sharp traders beat a million casual ones. For ad targeting and AI training, the ranking flips: a billion noisy-but-resolved forecasts, each one a labeled data point, are worth more than a small pool of precise ones. Arena doesn't have to produce good prices. It has to produce a lot of graded human predictions, attached to identities Meta already monetizes.
Which is the whole tell. Meta isn't really trying to get into the betting business. It's trying to manufacture, at planetary scale and for free, the one signal social media never managed to capture, what its users actually believe is going to happen, and feed it to the two machines that pay Meta's bills.