668 Addresses
The prediction market success story is also its extraction mechanism.
Polymarket has 1.7 million trading addresses. Of those, 668 — fewer than 0.04 percent — captured 71 percent of all realized gains. That is $3.7 billion in profit, concentrated in a population that would fit inside a mid-size lecture hall.
At the other end: 70 percent of addresses lost money. Only 0.51 percent of all wallets have realized profits exceeding $1,000. The top 0.23 percent of wallets controlled 63 percent of all-time volume — $20.2 billion. The bottom 59 percent of wallets accounted for 1 percent of volume.
Power-law distributions are not unusual. Equity markets, options markets, venture capital, and poker all exhibit similar concentration. What is unusual is the distance between this distribution and the narrative used to justify these platforms and their expansion into the financial lives of millions of people.
The structure of extraction
The distribution is worth examining in detail because of what it reveals about how the value moves.
At the top, 668 addresses with profits exceeding $1 million collectively hold 71 percent of all realized gains. Below them, 2,551 addresses earned between $100,000 and $1 million. Below that, the largest group of profitable traders — 24.56 percent of addresses — earned between $0 and $1,000, accounting for 0.86 percent of total profits. The structure is not a bell curve with fat tails. It is a funnel where capital enters at the bottom and exits at the top.
Academic research from the IMDEA Networks Institute, covering 86 million transactions across 17,218 market conditions, confirmed the zero-sum structure and went further. Arbitrageurs alone extracted more than $40 million in risk-free profits. The top three arbitrageur wallets made $4.2 million. The common traits of top traders were not predictive ability. They were: systematically capturing pricing errors, disciplined risk management, and patience to build information advantage in a single domain.
The anonymous trader RN1 illustrates the mechanism. He turned roughly $1,000 into $1.9 million through more than 2,000 trades. His strategy never involved predicting the future. He exploited three microstructure features: synthetic sells to avoid taker fees in illiquid order books, purchases of near-zero contracts to generate notional volume and capture platform rewards, and rebalancing inefficiencies where he provided liquidity at irrational prices. RN1 is not an information producer. He is a structural extractor. And the platform's incentive design made his strategy optimal.
This is not wisdom-of-crowds aggregation. It is market microstructure exploitation by a small number of sophisticated actors, funded by a large number of unsophisticated ones.
The reflexive loop
The standard observation — most retail traders lose money — could be written about any market in any era. What makes prediction markets structurally different is the feedback loop between the success narrative and the capital supply.
The mechanism works as follows. Prediction market platforms are valued at more than $20 billion. That valuation depends on growth. Growth depends on new users. New users are acquired through a narrative. The narrative requires success stories.
When NPR profiles a 25-year-old making $100,000 a month trading prediction markets from his laptop, it functions as a marketing event for the entire industry. The story circulates through Reddit, Discord, X. It reaches the demographic these platforms are designed to capture: young, male, high risk tolerance, adjacent to crypto or sports betting culture. Logan Sudeith, the trader NPR profiled, is a skilled practitioner. He quit a $75,000-a-year job, works 100-hour weeks, and his biggest single win was $40,236 on Time's Person of the Year. His edge, by his own account, is identifying mispriced probabilities.
Sudeith is a real trader with a real edge. But his profile does not describe the modal experience. On the same platforms and subreddits where his story circulates, a Reddit user named Educational_Pain_407 posted: “down 2000 this week...don't have enough to pay my bills.” There is no NPR feature on Educational_Pain_407. They are a data point in the 70 percent — one of roughly 1.2 million addresses that have lost money on Polymarket.
The loop closes here. The success stories attract new retail participants. New retail participants become the liquidity pool. The liquidity pool enables the next round of profits for the 668 addresses. Those profits generate the next round of success stories. The narrative of democratization is the mechanism of extraction.
On a zero-sum platform with minimal fees, every dollar of profit requires a corresponding dollar of loss. The $3.7 billion captured by the top 0.04 percent was supplied by the bottom 70 percent. The supply is not self-sustaining. It requires continuous recruitment. And the recruitment depends on the narrative.
Information as public good, losses as private cost
The “info finance” thesis makes an implicit empirical claim: if these markets produce valuable public information, the value is broadly distributed. Information is non-excludable. When a prediction market produces an accurate probability estimate, that estimate is available to everyone.
In practice, this is exactly what happens — but only on the information side. Google surfaces prediction market probabilities in search results. CNN broadcasts them. The Wall Street Journal publishes them alongside earnings calendars. The information benefit is free and universal.
The monetary cost of producing that information — the losses required to generate the liquidity in which prices form — is borne almost entirely by the least sophisticated participants. The information is a public good. The losses are a private cost. And the private cost falls on the people least equipped to bear it.
This is not an inherent flaw in the concept of prediction markets. It is a consequence of specific design choices. No KYC, zero fees, one-click deposits, push notifications, mobile-first interfaces, student ambassador programs — these features maximize the flow of unsophisticated capital into a zero-sum environment. Kalshi ran a college recruitment campaign explicitly designed to onboard students as users. The page was taken down after backlash. Three active class-action lawsuits allege these platforms “sucked young traders into gambling addiction.” The target demographic is not accidental.
The question of insider trading is separate but compounds the problem. When a brand-new Polymarket account wagered $32,000 that Maduro would be removed from office hours before Delta Force commandos captured him — and walked away with $400,000 — the platform had no mechanism to prevent it, detect it in advance, or compensate the traders on the other side. Evan Semet, one of the most profitable traders on Kalshi, runs statistical models on a dedicated AWS server and trades six figures monthly. Asked directly whether the platforms are doing enough to combat insider trading, his answer: “Fuck no.”
What 668 means for the industry
The prediction market industry in early 2026 is growing at an extraordinary rate. It has partnerships with the NHL, UFC, MLS, the Golden Globes, Google, the Wall Street Journal, and CNN. It is ranked number one on the App Store. The CFTC has one of five commissioner seats filled.
None of that growth changes the distribution. The 668 addresses exist because the architecture permits them to. Prediction markets produce a genuine public good — real-time probability estimates that are, on important questions, more accurate than any alternative. But the production mechanism is a zero-sum market where the cost of that public good is extracted from the participants least able to identify what they are paying for.
Every success story that enters the narrative attracts capital. That capital becomes liquidity. That liquidity becomes the next set of gains for the addresses at the top. The gains become the next success story. The loop does not resolve. It scales.
This is what a reflexive system looks like from the inside: a market whose growth mechanism is indistinguishable from its extraction mechanism, where the price of information is paid by the people who do not know they are paying it.