by Theo Encrypted

When does crypto gambling end: pump.fun vs Fomo

Almost half of the money spent on a pump.fun token arrives in the first thirty seconds of its life. The only participants who finish in profit are the ones who pressed create. Between 40 and 68 percent of what is lost is taken by the venue itself.

People ask this constantly. We ran the numbers and got an answer we did not expect.

The headline figure: we took every dollar spent buying pump.fun tokens during the first week of June 2026 and sorted it by time since the token was created. 46.4 percent of volume lands in the first thirty seconds, 77.5 percent in the first six minutes. The median token launched in July lives 3.39 minutes from creation to final trade.

You cannot evaluate an idea in thirty seconds. So what is being bought is not the idea but the speed: whoever got there first wins.

How we counted

Two cohorts, each with a full thirty days of observation. The first: every token created between 1 and 7 June 2026, 191,192 of them. The second: created between 6 and 12 January 2025, at the peak of the mania, 373,603. Separately we took all of July 2026: 857,288 tokens, 1,281,658 wallets, 77.7 million trades. This is a census, not a sample. Both cohorts were fixed by creation date in advance and were not picked to fit a result.

All money conclusions are limited to trading on the bonding curve, the mechanism pump.fun uses to sell a token before it reaches an exchange. Once a token graduates off the curve it moves to PumpSwap, which is pump.fun's own exchange. That is the second half of the same product, not some third party. Its data is excluded, and the reason sits in the "What we cannot see" section: the flows there do not physically balance.

The clock

Money does not arrive once something is known about a token. It arrives the moment the token exists.

One more number: 89.7 percent of cohort tokens received a buy in the same slot they were created in. In January 2025 that figure was 98.4 percent. The token gets bought in the same second it gets made.

The graveyard

Almost nothing that gets created survives to the next day.

The July cohort, where every token has a full month of observation, gives the same answer from another angle: one token in 203 is still trading thirty days later.

Averages are useless here. Half the tokens have no more than 11 trades, the mean is 58.8, and the maximum is 49,033. A handful of survivors drags the mean up, so "the average token" describes nothing at all.

Who wins

Sort buyers by how many seconds after token creation they made their first purchase.

The January 2025 data says the same thing: profit in the first group, losses in every other.

And here is the key line in that table. Of the first group's profit, 84.3 percent went to the token creators themselves. Everyone else who managed to be there at the same moment split five times less across far more trades. Being first here has less to do with speed than with having pressed create.

The house

"Zero-sum game" means one person's win is another person's loss. Cohort turnover on the curve was 5,846,461 SOL. The published pump.fun curve fee is 1.25 percent, which is roughly 73,081 SOL.

The source documentation does not say whether the fee is inside the recorded volume, so we compute both. The conclusion holds either way: between 40 and 68 percent of everything participants lose on the curve is taken by the venue, not by another participant.

The sum is not zero, it is negative. Part of the money leaves the game entirely, and that part is larger than what the players take off each other. They are playing the house more than they are playing each other, and the house takes over half.

The same arithmetic by wallet, for July:

Among wallets that made anything at all, the bottom ninety percent split two percent of the total profit. And "they only lose on fees" is wrong: fees are a little over a quarter of the losses.

What it costs, and who gets it

Annual pump.fun revenue is roughly 280 million dollars. That is a lot: four times Uniswap V3 and six times Aave V3.

The ratio also depends on what you compare against. The pump.fun ecosystem collects 65.2 percent of everything Solana validators earn from blockspace and extractable value combined. Against Solana, the word "parasite" is defensible. Against all crypto fee revenue it is 2.02 percent, and against the gross gaming revenue of US commercial casinos, 0.79 percent. The same phenomenon is a measurement under one denominator and rhetoric under another. The denominator has to be named every time.

Where the money goes

This is where it gets more interesting, because the venue does not keep its revenue.

The mechanism is simple. A trader pays a fee, the fee becomes a token purchase, the purchase supports the price of the token held by holders. One layer of the market funds another.

It works badly. The token listed at $0.004 in July 2025, raised 600 million in twelve minutes and 1.32 billion in total. Today it trades at $0.00273, which is 32 percent below the listing price and 69 percent below its all-time high. Three hundred and twenty million dollars of buying did not change that.

In April 2026 the share of revenue going to buybacks dropped from around ninety percent to forty-eight. The token then made a low in June and came back. The data shows no relationship between buyback size and price.

Who is behind the other side

We know more about the money on the other side of this argument, because that side files with a regulator.

Fomo Labs Inc. was incorporated in New York in 2024, SEC identifier 0002097616. It raised in three rounds:

The Form D filed on 3 June 2026 declares an offering of $73,999,844, of which $67,324,492 was actually received from nineteen investors. Related persons include Chetan Puttagunta of Benchmark and Julia André of Index.

One correction to how that number usually gets retold. Ninety-four million is not a round, it is the sum of three rounds since 2024. And the 550 million dollar valuation appears in no filing at all: a Form D does not report valuation. The figure comes from a Fortune piece on 22 June, and the founder confirmed it on a podcast five days later. We print it as claimed, not as established.

Fomo itself is real as a business: 8.2 million in revenue over thirty days, sixteenth in the world among protocols. It has no contracts of its own and routes orders into other Solana venues.

The user numbers, though, did not survive checking. The platform cited a hundred thousand daily users on 16 July and a hundred and thirty thousand on 24 July. We could not find a primary source for either figure: not in the company blog, not in its feed, not from the founders. What we did find were ceilings. Forbes reports roughly 650 thousand signups with about a third converting to traders, so somewhere near 195 thousand people have ever traded. A hundred and thirty thousand daily would mean two thirds of every trader in the platform's history trades every single day. The app store, meanwhile, shows "100,000+ users", which is a cumulative figure and looks like the source of the first number.

What a follower actually pays

Fomo has a mechanism worth showing on its own, because no retelling of the story describes it.

So a follower copying a trader who sends five-thousand-dollar orders pays nine and a half times the percentage the person they are following pays. Before the price moves at all.

There is a separate oddity here. Fomo's own help centre says, word for word, "fomo does not have a copy trading feature". Yet the site description served on every route promises "Follow traders, copy trades", and eight separate pages explain how copy trading works there. The disclaimer is dated May, the marketing August.

Who these participants are

Everywhere above, "participant" means a wallet, not a person. That matters: Solana wallets are free, and some of them act faster than any human can.

Two hundred and forty-nine wallets traded more than a thousand distinct tokens in seven days and did 4.4 million trades doing it. People do not trade like that. Under two percent of addresses account for more than half of turnover.

The "hundred tokens a week" threshold is ours, and it is wrong in both directions. It catches ordinary arbitrageurs as well as bots, which pushes the estimate up. It misses narrowly focused bots and networks of many small wallets, which pushes it down. Which effect dominates, we do not know.

We also tested the other theory, that most of this volume is fake. The clearest signature is one wallet buying and immediately selling the same token. That accounts for 0.76 percent of volume. But that is only the crudest case. An academic paper counting differently reports 21.4 percent. The gap is what our method cannot see: trades between two wallets, and trades with a delay. Printing 0.76 without that caveat would be dishonest.

Creators

There is a view that the game is rigged for whoever creates the token. The data supports that halfway, and the other half is inconvenient for both sides of the argument.

For "rigged": eighty-five percent of creators buy their own token and seventy-eight later sell it. People do that on purpose, and they do it at scale.

Against: the size of the haul. The median creator address earns zero. The single largest week's take across 191 thousand launches was twenty-five thousand dollars. All creators together pulled 4.58 million against the 209 million that entered the cohort as purchases, so roughly two percent. "Creators are fleecing people" does not survive the arithmetic. They do fleece, for pennies, and the real money is lost elsewhere.

Now the caveat that could overturn that conclusion. We measured only the creator's own address. If a creator buys through other wallets they control, we do not see it, because seeing it would require tracing funding transfers between wallets and we did not do that. So the estimate is low by an unknown amount. There is a hint of it directly in the data: 41,624 tokens were bought at the moment of creation, but only 8,453 distinct buyers did that buying. A small group of wallets makes the first purchase across a great many other people's launches. Whether those are launch services or somebody's wallet networks, we could not separate.

What is clearly visible: launching tokens is an industrial process. 275 creators, seven tenths of one percent of the total, produced 40.3 percent of the cohort. Ten wallets made 18,328 tokens in a week, 2,618 a day between them. Their first-purchase sizes repeat to the decimal place. That is a program buying, not a person.

The argument

The people arguing are known by name in this market. Worth noting: they barely replied to each other, so these are two separate positions rather than a conversation.

Ansem argues that what is happening is fine. Responding to a post about Pump and Fomo competing, he writes:

2026 and 2027 will produce a lot of self-made people from trading onchain and perps on equities and majors. Previously these people were not public, now it will happen in social mobile apps in front of a large audience, and virality will grow the industry a hundredfold.

He does not dispute how the machine works, only the verdict on it: trading, in his words, becomes "the most popular player-versus-player game for the next generation". He discloses his own stake himself: Ansem owns the Bullpen platform and the ANSEM token.

On the other side is CryptoCred, 815 thousand followers, with a mechanism stated in a way you can test:

Broad altseason is an artefact of the past and very hard to repeat, because there are too many coins now, and the excess speculation no longer happens on centralised exchanges: it has been siphoned into pairs on maximally PvP terms.

The same person on what the entry point has become for a retail participant:

I'm not sure bringing back the trenches is a worthy goal, given they were one of the most efficient retail meat grinders we have seen in a long time.

The most-read line belongs to cobie, at 1.2 million views:

A nice feature of trading metals is that you don't have to worry about copper inu stealing all the attention from copper while you sleep. New commodities don't get launched on pump.fun every few seconds.

The unexpected hit came from inside the memecoin camp. Murad, 722 thousand followers, the day before the August pump.fun announcement:

99% of crypto twitter wrongly believes attention is the main thing. Wrong. The main thing is fanatical belief across multiple cycles. The first gives you a one-time pump. The second sets you up for life.

He is not arguing against memecoins. He is arguing against the business of selling attention and turnover, which breaks the usual "memecoin people versus everyone else" split.

What the other side says

Se Yong Park, co-founder of Fomo, gave a ninety-minute interview to The Rollup. We pulled the full transcript, 22,175 words, and here are three places where he describes in his own words what we measured in numbers.

First: the newcomer's journey ends at the first token.

Our own wallet measurement on pump.fun puts a number on the same thing: 96 percent of wallets that arrive for the first time in a given week do not appear the following month. The founder of a competing venue describes this disease more precisely than any outside critic, and is building a product as the cure for it.

Second: he calls the game a win-lose game outright, and treats that as the moat.

The argument rests on the sum being zero: one person is right exactly as much as another is wrong. Our arithmetic says otherwise: the fee makes the sum negative and eats 40 to 68 percent of what is lost. The right and the wrong are left splitting less than half.

Third, and this is the most awkward one for his own product.

We tried to verify that from the outside and could not. The Fomo leaderboard has no public route: the address /leaderboard returns the same app shell every other address does, and it is absent from the client route table entirely. Programmatic access needs a key. The web archive holds no snapshot of the board. Follower counts are visible only inside the app.

So whether "the person who has been right more often" stays right from month to month cannot be checked by anyone except the company. Blockchain transparency is beside the point here: the trades really are visible, but the ranking built from them, and the rules for building it, are not. Those are different things.

He also states the leaderboard rule this way:

It is a good rule, but the platform's own help centre says unrealised profit counts toward the ranking, and there is no minimum trade count at all. So the top can be reached by someone holding paper profit in an illiquid token they cannot exit. And they can exit it, if their followers buy.

The second co-founder, Paul Erlanger, describes the company's goal as broader than trading: becoming one of the largest media platforms in the world in order to support its creators. That is an honest description of the business, and it explains why both venues currently pay for attention rather than for execution.

On the money specifically. The Rollup reports that Fomo hit weekly revenue of 2.23 million dollars, an all-time high, and quotes the founder saying "we're not even squeezing revenue" and "I'd trade ten years of revenue for a hundred percent month-over-month growth". Those words are not in the interview transcript, we checked five phrasings, so we attribute them to The Rollup's post rather than to the video.

The weekly 2.23 million does line up with our independent measurement: 8.2 million over thirty days per DefiLlama. The "not squeezing revenue" position sits awkwardly next to a twenty dollar order paying 9.5 percent round trip because of the fee floor.

The market's bet: PMF over rotation

While the founders argue about platform design, the market votes with themes. DeFi analyst Ignas bets that the defining crypto narrative from 2026 onwards is PMF — product-market fit: products with measurable demand instead of the pretty stories the market traded on for years.

His list of products with real demand covers protocols with high fee or holder revenue, stablecoins, prediction markets, perps, tokenization — and trading apps named outright: Fomo and pump.fun. Both platforms of this article pass his filter. The revenue we counted above — $320.8 million routed into buybacks in a year at pump.fun and $2.23 million of weekly revenue at Fomo — is that PMF, just measured.

Attached to the post is a Tiger Research table: twelve months of 2025, twelve narratives replacing one another.

The desk's read: the rotation we found inside pump.fun runs one level up as well. There, tokens replace each other in minutes; here, market narratives do it in months. In both cases attention moves on to the next object and the previous one stops receiving new capital.

If Ignas is right and in 2026 the market really does start pricing products by fee revenue rather than by the next story, pump.fun and Fomo only gain from it. They do not need to guess which token or narrative wins next. They earn on the act of trading itself, and the numbers above show how much.

Hard to prove, yet everyone understands it

Now the uncomfortable part, and it is about us.

We started from this idea: memecoins pull money away from the rest of the market, so altcoins do not rise and trading becomes a game nobody wins. The second half held, and turned out worse than we thought. The first half failed all four tests.

Correlation. Twenty-eight months. Altcoin breadth was computed from the desk's own point-in-time store, including delisted pairs, so we were not looking at survivors alone. The hypothesis predicts a negative relationship. All six measured coefficients came out positive, from +0.145 to +0.243, and not one reached significance. The control variable, bitcoin return, explains exactly as much.

Historical control. If memecoins are the cause, altcoins should do better without them.

The worst stretch for altcoins in the current cycle ran from late 2022 through all of 2023, when bitcoin dominance climbed from 38.6 to 50.9 percent. pump.fun did not exist yet, it launched in January 2024. And the strongest altseason of the modern era coincided with the DOGE mania. Through the pump.fun era itself, altcoins tracked bitcoin.

The money. If memecoins drain capital, stablecoins should leave the chain during a mania. Stablecoin supply on Solana went from 5.12 billion in December 2024 to 11.43 billion during the single month of mania, and it stayed: 16.31 billion today, three times the pre-peak level, even though meme volumes have since collapsed.

A natural experiment. The AI agent sector on Base and Solana ran a complete cycle: a peak of 9.70 billion in December 2024, 1.31 billion now, down 86.5 percent. The sector died. Did the liquidity come free?

Breadth did not rise. It fell by two thirds. Partial correlation controlling for bitcoin market cap came out at +0.526, significant at better than five percent. The sign is the opposite of what the hypothesis predicts, and this time the result is significant.

We will name the weakness ourselves. The agent sector died during a general market decline, and "everything fell together" is also an explanation we cannot rule out. Both numbers may simply reflect how much appetite for risk there was. So the precise statement is this: the test does not prove causation, but it finds no drain either, and what it does find points the other way.

Robinhood built a permissionless chain

While everyone argued about where all this should happen, the answer arrived from a direction nobody was watching.

On 1 July 2026 Robinhood, a listed broker with 28.4 million funded customers, launched its own chain on the Arbitrum stack. It was announced for tokenized equities. Forty days later:

The launchpad names say it: RobinFun, HoodPump, HoodMint, StonkBrokers, StockRip, ArrowPad.

The regulated product the chain was announced for lives on a different chain. Robinhood's tokenized equities are tracked on Arbitrum One: $14.0 million in assets and one trade, for $1.27, across two hundred days of observation.

Robinhood's own documentation states plainly that anyone can publish on the chain. The market took the broker at its word within six weeks.

What we cannot see

• The second half of the product is excluded. PumpSwap, where a token moves after the curve, belongs to pump.fun itself. Its inbound and outbound flows leave an unexplained gap of 2.53 million SOL that survives a duplicate check, so we left it out of the arithmetic. That means our totals describe only the first stretch of a token's life: everything earned and lost after graduation is not in them. Any public dashboard that adds the curve and PumpSwap together inherits that gap invisibly.
• A wallet is not a person. Every per-participant figure is computed per address. Wallet rotation understates retention and overstates the count of small losers. We did not size the correction.
• Creator funding wallets were not traced. The "median creator earns zero" conclusion may not survive a full transfer trace.
• Fomo's internal data is closed. The leaderboard has no public route and no archive snapshot, the interface is key-gated, and follower counts are visible only inside the app. The historical state of the ranking cannot be reconstructed from outside, so there is no way to test whether "top traders" persist month to month.
• No academic wallet-level profit and loss distribution exists for pump.fun. The largest dataset, covering 2.6 million traders, does not report the share who lost. Every "X percent lose" figure circulating in the market traces back to dashboards through press retelling.
• Nobody has measured what share of losses fits an entertainment budget. This is the one counterargument cohort data does not address: some participants pay to play knowingly, the way they pay for a cinema ticket.

What would have to be true for us to be wrong

The claim of this note: almost all the money arrives in the first seconds of a token's life, the net gain goes almost entirely to whoever pressed create, and a large share of the aggregate loss is taken by the venue.

It is refuted if any of the following holds:

1. The "creation slot" group stops being the only profitable one on new cohorts. So far it has held twice, seventeen months apart. 2. The fee turns out to be less than a quarter of aggregate curve losses. Settled by documenting whether the source records volume before or after the fee. Settleable, but not settled. 3. Tracing funding wallets shows creators take the bulk rather than crumbs. Then the "rigged market" reading wins, and our zero-median finding turns out to be an artefact of a narrow definition. 4. Retention corrected for address rotation comes out an order of magnitude above the four percent we measured.

Corrections we made along the way

We publish these because all three pointed toward a louder claim than the data supports.

The phrase "zero-sum market" was ours, and it was wrong. The sum is negative by the size of the fee, and the fee is 40 to 68 percent of losses. We described the game as gentler than it is.

The claim that "pump.fun cut its fee", under which the August announcement was presented, does not hold in a single day of data. The median curve rate was 95 basis points every day before and after the announcement, and venue revenue rose afterwards. More than that: the effective rate over the year went from 0.47 to 1.30 percent.

The top-three-by-revenue claim we checked and could not confirm. Comparing all protocols on the same basis, the venue is sixth.

What this actually is

Back to the question we started with.

The discovery layer gets described two ways, and both are inaccurate. The defence says it is a new form of distribution, a shop window where an idea meets an audience. The prosecution says it is a casino draining the market of attention and money.

The data supports neither description in full. It is not a shop window: eighty-eight percent of what goes on display is gone within an hour, and almost half the money arrives before the idea can even be read. But it is not a casino in the sense charged either. It does not drain the wider market, four independent tests found no such channel, and the stablecoins that arrived on the mania stayed on the chain after it ended.

What is very visible instead: what matters here is less what was bought than the moment of buying. The only group that finished in profit entered in the slot the token was created in. Six minutes later the odds of a profitable trade are already down to 28 percent. And 84.3 percent of the first group's gain sits on the creators' own addresses.

So when does it end? Judging by our measurements, the question needs rewording. What ends is not "gambling", because gambling takes nothing from the market, contrary to the common charge. What ends, if anything does, is the advantage of the first four hundred milliseconds, because that is what turns trading into a lottery with the winner known in advance. And that is a question about queue design, not morality.

The desk's view

What follows is about us rather than the market: what we do differently after these measurements.

We are retiring one of our own claims. The desk held that memecoins pull liquidity out of the rest of crypto and therefore postpone altseason. We went to check it and found no channel: not in correlation with the broader market, not in stablecoins, not in cross-chain flows, not in exchange volumes. Four independent tests, four empty results. The claim was ours, it did not hold, and it comes out of our market write-ups. "The casino is stealing money from altcoins" sounds convincing and measures as nothing.

We are changing what we look at on new tokens. Until now the desk looked at liquidity, pair age and buyer count in new listings. After these cohorts we add one more: how much time passed between token creation and the trade we are considering. If that is not the first few seconds, the historical share of profit going to such entries is close to zero, and the other attributes stop mattering. It never argues for a buy. It only ever argues for passing.

What we track from here. Two figures, both public, both computed by one query. First: how much of the winnings the creation slot keeps, currently 84.3 percent. Second: the venue's effective rate, currently 1.30 percent against 0.47 a year ago. The first tells us whether the window is narrowing. The second tells us whether participating is getting more expensive. While both rise, the story does not change, whatever gets announced.

And a request. Check one number behind us: of the first slot's winnings, 84.3 percent went to the wallets that created the token. The method repeats, sort participants by entry time. If that number holds on other venues and other chains, the conversation about the discovery layer stops being about attention or distribution and becomes a conversation about what happens in the first four hundred milliseconds.

*This is research, not investment advice. No number here is a forecast and none of it promises a return.*

Keep reading

Crypto Capital Flows Analysis · All alpha