Polymarket split into two markets in the first half of 2026
Surf said its findings are based on 787 million deduplicated trades and 1,733,011 wallets on Polymarket’s international market between January and June 2026, combined with market-level data from Polymarket US. Its central point is simple: Polymarket no longer functioned as a single venue over that period.
The international market, which serves non-US users, continued to run on an off-chain CLOB with settlement on Polygon. Over six months, it recorded $17.9 billion in one-sided volume. At the same time, Polymarket US, launched in late 2025 and regulated by the Commodity Futures Trading Commission, climbed from 5% of the international market’s monthly volume in January to 47% in June. In July, that ratio reached 67%, or about $2 billion versus $3 billion.
The report said the two venues share the same brand and a similar order-book design, but not the same order book, user base, or revenue mix. Like Kalshi, both use off-chain matching, a central limit order book, and the same price-sensitive taker fee curve: fee = Θ·C·p(1−p). The difference is in settlement. The international venue settles each trade publicly on Polygon, while the US market clears through a private regulated ledger.
Surf also standardized the definition of volume before making platform comparisons. In the report, volume means one-sided, taker-side, actual traded dollars counted once. Notional figures and double-sided counts are usually 2x to 3x higher. Reworked on that basis, the common July figures of $41 billion, $7 billion, and $5 billion correspond to about $13.3 billion, $3 billion, and $2 billion.
Polymarket US expanded quickly, with about $30 million in June gross taker fees
Daily trading on the US venue grew from roughly $1 million near launch to $50 million to $100 million in July, according to the report. Daily trade count moved past 1 million, and open interest at one point rose above $150 million. The fee model changed too: January used a fixed basis-point structure, and April brought in a price-sensitive formula.
Surf estimated that gross taker fees on Polymarket US reached roughly $30 million in June alone. Using the official rebate program as a guide, the report assumed 25% was returned to makers at execution, noted that a full taker promotional rebate ended on April 30, and said later rebates became volume-tiered. On that basis, June retained revenue was about $17 million, around twice the figure produced by many trackers that assume a flat 25% rebate.
The US venue does not offer user-level data, so the report could not directly map large taker tiers there. Instead, it pointed to the international market, where the top 0.3% of takers accounted for about 63% of volume.
Sports made up about half of international volume, while crypto accounted for roughly 30%
On the international market, sports represented about half of all volume in the first half. Crypto made up roughly 30%, most of it from five-minute up-or-down micro markets. World and geopolitics accounted for about 15%.
Concentration varied sharply by category. In sports, the largest market represented only 0.27% of category volume. In science and technology, one market — “Will the US confirm the existence of aliens before 2027?” — made up about one-quarter of the category.
The international market generated about $184 million in gross fees in the first half of 2026, though Surf said that figure was before maker rebates and incentives. Fee production was concentrated in the fastest-turning categories. Crypto generated about $96.3 million in fees, more than sports at about $83.5 million, despite doing only about half as much volume. Together, those two categories contributed 98% of total fees. World and geopolitics ran on an intentional zero-fee policy.
That led Surf to a broader conclusion: high-turnover categories are the real fee engines.
Turnover looked more like a category trait than a platform trait
Surf compared Kalshi, Polymarket international, and Polymarket US and argued that capital size alone did not explain trading speed.
In July, Kalshi’s open interest was about 2.3x that of Polymarket international, but its volume was about 4.4x higher, translating into a daily turnover rate of roughly 0.44x versus 0.23x. Polymarket US had the smallest capital pool, yet posted the fastest daily turnover at about 0.64x.
At the category level, the pattern was clearer. Sports and crypto were the fastest-turning categories. World and geopolitics were the slowest, at only about 0.01x to 0.03x per day. Two international venues each had roughly $200 million to $300 million tied up in that category, but produced only $2 million to $3 million in daily volume, which is why both kept it at zero fees.
About 88% of Polymarket US open interest sat in sports, which helps explain why it achieved higher turnover with less capital. Kalshi’s crypto markets were a clear outlier: about $17 million in open interest against around $96 million in daily volume, close to 6x daily turnover, with 98% of those markets expiring within one hour.
Polymarket’s crypto turnover looked lower at about 0.31x, but Surf said that was partly an accounting effect. Long-dated threshold markets carried most of the open interest, while five-minute price-direction markets accounted for about 90% of volume and locked up very little capital. On average, each such market tied up only about $2,800 in collateral.
The same category also tended to trade at similar speeds across venues. Sports turned over at about 0.57x on Kalshi, 0.67x on Polymarket international, and 0.64x on Polymarket US. Surf said that looked more like a category attribute than a platform-specific one.
1.73 million wallets traded, but monthly active wallets were almost flat
The user analysis covered only the international market’s Polygon CTF and NegRisk contracts from Jan. 1 through June 30, 2026. Surf noted that Polymarket US and Kalshi settle on internal ledgers and release anonymized public trade flows, so equally deep wallet-level profiling is not possible there.
A total of 1,733,011 wallets traded on the international market in the first half. Almost every wallet took liquidity at least once, but fewer than half ever posted a limit order. The maker side was smaller and more specialized, with the top 1,000 makers supplying most static liquidity. Demand was concentrated too: about 6% of highly active wallets generated 80% of taker volume, while one-third of wallets traded five times or fewer over the six-month period.
User growth looked weaker than headline volume. Monthly active wallets moved from 580,000 in January to 594,000 in June, essentially flat, even though about 1.1 million wallets traded for the first time during the same period. Each month, 43% to 49% of the prior month’s active wallets stopped trading. In April, churn reached 358,000 wallets, forcing the platform to refill the funnel with new and returning wallets.
Surf added three caveats. A wallet is not the same thing as a user. Some of the apparent churn may reflect users rotating into fresh wallets. And in this report, churn only means a wallet stopped trading, not that an account was closed or had no funds left.
User acquisition did not come from a single breakout market
When Surf attributed each new wallet to the first market it touched, the result was a seasonal but dispersed acquisition pattern. Crypto brought in about 99,000 new wallets in March. Geopolitics brought in about 68,000 in January. Sports drew about 91,000 in June as the World Cup began.
The strongest single acquisition market in the first half was “Will China invade Taiwan before the end of 2026?” Even that market brought in fewer than 7,000 new wallets. The top ten included geopolitics, World Cup markets, rates, and cultural events. The report said effective acquisition came from a mix of markets rather than one blockbuster.
Viewed month by month, the top 15 acquisition markets accounted for only 8% to 19% of new wallets, and the number one market never contributed more than about 3% in any month. The highest concentration came in January and June, linked to Iran-related news, Taiwan markets, and the World Cup.
Platform retention held near 51% to 57%, but category loyalty was weaker
About half of monthly active wallets returned the next month, putting platform-level monthly retention at roughly 51% to 57%. Category loyalty was noticeably weaker. Many users left the category they entered through and reappeared somewhere else on the platform.
Across most entry categories, one-month retention clustered around 44% to 49%, while three-month retention was about 25% to 32%. Crypto was the clear low point. Wallets entering through crypto posted roughly 37% one-month retention, about 30% lower than other categories. Surf offered one possible explanation: automated activity in crypto may rotate through fresh wallets more often, pushing down apparent retention.
Cross-category participation also showed a hierarchy. World and geopolitics overlapped more heavily with other categories, except weather. Sports and crypto were the two biggest user groups, but they overlapped by only 49%, suggesting two meaningfully different audiences. Surf described world and geopolitics as a connective layer across the platform.
Broader category coverage and higher frequency usually meant larger trading size
Surf profiled users through three lenses: how many categories they touched, how large each trade was, and how often they traded. Wallets active across more categories had higher average volume. Wallets that traded all seven categories made up only about 1% of users, but averaged roughly $59,000 in taker volume in the first half, seven times the $8,300 seen in single-category wallets.
Frequency followed a classic 80/20 pattern. About 6% of high-frequency users generated 79% of volume, while tourist-like wallets that traded only once made up about 8% of the user base.
In most categories, a wallet’s relationship with a market was defined by one or two trades. Sports showed the highest share of one-trade relationships at 46%. World and geopolitics looked different: only 20% of wallet-market relationships were one trade, suggesting users kept returning to the same market as news evolved.
Retail wallets were about 90% of users, but bot wallets handled about two-thirds of volume
Surf grouped wallets into bots, professional or informed users, and retail users using behavioral heuristics such as trading frequency, consistency of order size, breadth of coverage, and average ticket size. It said the labels were not claims about true identity and should be read as a way to compare trading style.
Bot-like wallets made up about 5.9% of users, yet accounted for about two-thirds of trading volume. Roughly 102,000 bot wallets generated $11.9 billion in volume and 89% of trade events. About 1.54 million retail wallets generated $4.1 billion. Professional or informed users accounted for about 5% of wallets and 6.6% of taker volume, fitting a profile in which they provide more liquidity and cross the spread less often.
By category, bots almost fully dominated crypto and sports, taking about 80% and 72% of volume, respectively. In geopolitics, finance, and culture, the shares were closer, leaving more room for human traders to influence pricing.
That volume advantage still came from a thin automation layer. On a wallet-count basis, every category remained overwhelmingly retail. Surf noted that crypto and weather had visibly higher bot penetration, with about 5% of population share replaced by automated behavior.
After fees were introduced, professional traders lost volume share more quickly
Because categories turned on fees at different times, while world and geopolitics stayed at zero fees, Surf treated the shift as a natural experiment. The report said volume mix changed materially after fees went live, especially in crypto, culture, and weather.
In those categories, professional or informed users lost volume share quickly, while retail and bots stayed comparatively stable, showing stronger fee insensitivity. Changes in wallet share were smaller than changes in volume share, but the direction was the same. In Surf’s reading, fees first changed who was sizing up, not necessarily who was still present.
Most users entered late, with sports and crypto showing the heaviest adverse selection
Looking at each market’s lifecycle, late entry was the norm. The degree of lateness varied by category. Sports and crypto were the most delayed and competitive: median entry for each group landed at 92% to 95% of the market lifecycle, creating stronger adverse-selection pressure for retail traders. Culture sat at about 85% to 90%.
Weather was different. Entry came earlier, with retail median entry around 61% and bots around 51%, in line with outcome-driven markets that can be priced earlier by models. Weather was also the category where professional users arrived relatively latest, with a median of about 83%.
This analysis included only markets with a fully observable lifecycle, meaning they settled in the first half of 2026 and wallet activity could be tracked from the first trade, including positions opened before 2026.
Surf also said that in the largest markets by volume across categories, retail flow tended to stretch across the whole lifecycle. Professional or informed users and bots were more likely to surge when information arrived or to concentrate bets near settlement.
Bots posted an aggregate $108 million net gain, but only 146 wallets stayed cash-flow positive for six straight months
For profitability, Surf summed cash flows from positions in markets that settled in the first half. Before fees, the ledger was close to zero overall. Of about 1.97 million wallets with settled positions, 34% finished positive. Big winners together made about $666 million, while big losers together lost about $530 million. Across all groups, the total was near zero.
Surf said that confirmed the calculation was gross PnL before fees. Once the first-half fee take of about $184 million is included, aggregate results would turn negative.
Broken down by cohort, the flow of money was clearer. Retail and professional users together served as counterparties to the bot cohort’s gains. Bots were the only group with positive net results overall, at about $108 million, but only 27% of bot wallets were profitable. The median bot wallet still lost money. Profit, the report said, was concentrated in a small number of large-scale operators, and persistent profitability was rare.
Professional users had the highest hit rate, with about 42% of wallets positive. Retail was about 35%, and bots about 27%. Even so, professionals as a group still lost about $26 million. Surf cautioned that the calculation did not include rebates, incentives, or other programmatic subsidies, so it should not be used to conclude that professionals were net negative after all payments. Retail users lost about $74 million in aggregate.
Out of roughly 1.75 million wallets, only 146 maintained positive cash flow for six consecutive months. Even among wallets that were profitable at least once, many earned more than 90% of their profits in a single month. Surf’s conclusion was that profits in prediction markets looked more like occasional spikes than a stable income stream.
Methodology
Surf said the report’s calculations were provided by Surf AI. PnL covered only markets that settled in the first half of 2026 and measured realized results before fees from wallet cash flows in those markets. Maker rebates, promotional incentives, and other subsidy programs were excluded from gross PnL. The cohort labels were behavioral heuristics meant to compare trading styles, not definitive identification of wallet ownership.