Pump.fun stock tokens: inside a $6 billion meme coin market
On 9 September pump.fun turned on Custom Pairs, so a new meme coin can be priced in a tokenized stock instead of SOL. It is the last of five launchpads to switch this on, and one of the smallest. Robinhood Chain has been running the same idea since 12 August, and BNB Chain quietly carries more of it than Solana does.
01 · The trade you are actually making
Buying a joke coin now means holding tokenized NVIDIA first
On 9 September, a little before eight in the evening UTC, someone on Solana bought a meme coin and paid for it in tokenized S&P 500. The payment was not SOL and it was not USDC. It was a token that tracks the index.
That trade was the first of its kind on pump.fun, the launchpad most people mean when they say meme coin. The feature is called Custom Pairs, and it lets whoever launches a coin choose what the coin is priced in. Tesla, NVIDIA, the Nasdaq, gold. Within 12 hours there were hundreds of coins doing it.
Some vocabulary, once. A stock token is a token that tracks a real share, issued by a company that says it holds the share behind it. Backed sells them as xStocks, Binance as bStocks, Robinhood and Ondo under their own names. A launchpad is a site that mints a new coin and gives it an instant market. The quote asset is what that market prices the coin in, and what a buyer has to hold before they can buy. Change the quote asset from SOL to tokenized NVIDIA and the buyer is now making two bets at once.
Bitquery indexes every trade on Solana, Robinhood Chain and BNB Chain. We took 31 days of that record and pulled out every trade where one side was a meme coin and the other was a stock token, 43.3 million of them, then followed the money each wallet put in and took back out. We wanted to know whether pricing a coin in a stock changes how it behaves, and what happens to the person buying. Then we followed the money to whoever ends up with it.
One limit belongs up front. Our window opens on 11 August because that is as far back as the trade record goes at this depth. Nothing here is an all-time figure, and pump.fun's own numbers cover barely a day and a half.
02 · The race nobody announced
Five launchpads, four weeks, and pump.fun arrived last
Pump.fun got the attention. It was not first, and it is not the biggest. Robinhood Chain's Pons launchpad started pricing coins in Robinhood's own stock tokens on 12 August, and now carries roughly 28 times as many of them. Raydium's LaunchLab on Solana followed on 24 August. Flap on BNB Chain switched on in September. Four.meme, also on BNB Chain, has been doing it quietly the whole time and calls the category Stock Memes.
| Launchpad | Coins priced in a stock token, 31 days |
|---|---|
| Pons (Robinhood Chain) | 23,845 coins · 30.7% of what it launches · first on 08-12 |
| Flap (BNB Chain) | 3,172 coins · 1.4% of what it launches · first on 09-07 |
| Raydium LaunchLab (Solana) | 2,836 coins · 54.6% of what it launches · first on 08-24 |
| pump.fun (Solana) | 853 coins · 0.3% of what it launches · first on 09-09 |
| Four.meme (BNB Chain) | 234 coins · 1.9% of what it launches · first on 08-11 |
The share matters more than the count, because it says what creators choose when they have a choice. On Raydium LaunchLab four in five new coins are now priced in a stock. On Flap the share went from nothing to about half in three days.
Coins priced in a stock also survive their first hour better and reach a real exchange more often. On Pons the typical one lasts 14 minutes against 5, and on every one of the five launchpads a higher share graduates off the bonding curve. Hold that thought, because it does not mean what it sounds like.
03 · Nearly three quarters of buyers are down
$2.98 billion went in and $2.70 billion came back
All of the below is cash. Dollars actually paid in against dollars actually taken out, with nothing marked to market, for a reason we come back to in the method.
| Buyers, 11 August to 10 September | |
|---|---|
| Wallets that bought with money | 951,190 |
| Paid in | $2.98 billion |
| Took back out | $2.70 billion |
| Net | -$287.3 million, or 9.6% of everything paid in |
| Wallets that ended ahead | 26.7% |
| The median buyer | -$11.46 |
Nearly half a million wallets bought one of these coins and never sold a single unit of it. That group alone is down about $302 million. The wallets that did get out again finished ahead by $251 million between them, which is the whole reason the market keeps working: enough people escape to make it look survivable.
Losses barely move between issuers. Whether the ticket is priced in a Robinhood token, a Binance bStock, a Backed xStock or an Ondo wrapper, between a quarter and a third of wallets end ahead. That points at the mechanism rather than any one issuer. Depth is part of it, and the tokenized stocks themselves are thin enough that a $10,000 order moves them before you add a meme coin on top. All of it is visible in ordinary DEX trade data.
Now the part that undoes the survival finding. Compare like with like on the one launchpad that ran both kinds of coin side by side, at scale, all month, matching them on the day they launched:
| Pons, like for like | Priced in ETH or a stablecoin |
|---|---|
| Wallets ending ahead | 36.5% vs 34.9% |
| Trades never sold at all | 11.7% vs 14.9% |
| Lost, as a share of money paid in | 5.9% vs 7.7% |
A coin priced in a stock lives longer and reaches a real exchange more often. Its buyers still do worse. A longer life is more time to be sold into.
04 · The wallets on the other side
418 wallets took two fifths of the money
If buyers are down $287 million, someone is up. The counterparty is easy to find once you look for wallets that sold a coin without ever buying it: creator supply, dev allocations, airdrops. There were about 264,000 of them, and they took $268 million having paid in nothing at all. That is 93% of what the buyers lost, moving the other way.
It is concentrated. A small set of wallets shows up in thousands of separate coins.
| Wallets that sold without ever buying | How many coins each appeared in |
|---|---|
| 418 wallets | 100 or more coins each · took $112M, or 42% of all of it · the busiest appears in 9,647 |
| 2,149 wallets | 26 to 100 coins · $23M |
| 15,863 wallets | 6 to 25 coins · $34M |
| 70,667 wallets | 2 to 5 coins · $39M |
| 174,936 wallets | a single coin · $59M |
That top row is a business running at scale. The long tail underneath it is real and small. Money-flow tools make the shape obvious once the wallets are grouped.
05 · Ten seconds
The only entry window that makes money
Sort every trade by how long after the coin's first trade the buyer arrived, and the answer is brutal and clean.
Winning here is skill, and almost nobody has it. Split each active wallet's trades into a first and second half and the win rates correlate at about 0.5, so whoever is ahead tends to stay ahead. The spread of win rates is more than twice what luck alone would produce. Even so, only about 1 in 100 active wallets wins more than three quarters of the time.
The volume itself is real, which is worth establishing before anyone blames wash trading. Trades where the same wallet buys and sells the same coin inside a single transaction come to under 1% of trades and about 2% of volume, whether on Pumpswap or anywhere else. The volume is real people losing real money.
06 · The stock half does almost nothing
You are not getting equity exposure. You are getting a six-minute meme coin
Buying a meme coin priced in NVIDIA is two bets: the coin against NVIDIA, and NVIDIA against the dollar. Those two split apart exactly, and across 7.1 million closed round trips the second one barely registers.
| Closed round trips, split in two | |
|---|---|
| The meme coin half | +$293.5 million |
| The tokenized stock half | -$42.5 million |
| Trades where the stock half decides who wins | 1.00% |
| Typical time held | 6.2 minutes |
| How far NVIDIA moves in that time | 0.022% |
Six minutes is the whole story. A blue chip barely moves in six minutes, so the stock on the ticket works as a label rather than a holding. The stock side does cost money at the long end, about $23 million on trades held between a day and a week, and it still almost never picks the winner. The exception is Four.meme's tokenized Cea Industries, a small and jumpy stock that moves about 60 times more than the Nasdaq token does over the same holding time, and there the quote asset starts to matter. Price feeds make that easy to check.
07 · A market that ignores market hours
Four fifths of it trades while the exchange is shut
The assets these coins are priced in can only be bought or sold anywhere else for about 32 hours a week. The coins trade all of them.
Nobody here is tracking the opening bell. The same pattern turned up when we looked at Binance bStocks trading at 3 a.m., and it is sharper now that a meme coin is attached.
08 · What this is doing to the issuers
Meme coins are now two fifths of all tokenized stock trading
This is the finding that should travel furthest, and it is not about meme coins at all. Take every trade in the window that touches one of these 403 stock tokens, whichever side it sits on:
| All trading that touches a stock token, 31 days | |
|---|---|
| Against stablecoins and major crypto | $9.65 billion — the intended use |
| Against meme coins | $6.29 billion |
| Against other stock tokens | $391 million |
Two fifths of the on-chain turnover of tokenized equities is now meme coin settlement. For Backed, Ondo, Binance, Backpack and Robinhood that is a plurality of what their product actually gets used for, arriving without a launch or a press release. When we mapped Solana's xStocks market earlier this summer this channel was a footnote. On BNB Chain, where Four.meme mints its own stock tokens whenever a ticker starts trending, it is now the point.
09 · How we measured it
The universe, the guards, and what we threw away
A quote asset counts only if a stock-token issuer minted it: Robinhood Chain tokens, Binance bStocks, Backed xStocks, PreStocks, Backpack Securities, Ondo Global Markets, Four.meme's stock tokens, and Robinhood's pre-IPO and Korean listings. That gives 403 of them. Searching by name fails here, because it drags in meme coins called Stock Cat while missing Binance's bStocks completely. Those carry plain company names like NVIDIA Corp with no issuer marking, and they are 40% of this market.
Pairs were collected in both directions. Most venues record the security as the quote asset, and about 600,000 trades run the other way round, with the security as the base. Those sit in different pools and are not duplicates.
Four things were removed, and each would have produced a wrong number. Robinhood Chain's Uniswap v4 hook writes each swap's fee as two extra swap rows, so more than a quarter of that venue's rows are not trades; that also reveals the fee, which is 1.14% per swap and came to about $35 million over the month. Trades between two stock tokens are a real market, and a different one. A single pool carried 201 rows whose amount was the largest number a 64-bit counter can hold, a decoding artifact. And marking unsold bags at the last traded price produced $2.39 billion of imaginary profit, with one position showing $383 spent and a bag valued at $363 million, which is why nothing here is marked to market.
Two things we are not claiming. The buyer comparison in section three is one launchpad, because it is the only one with both kinds of coin running all month. And pump.fun's Custom Pairs are not stock-only; coins priced in PUMP, USDC and wrapped bitcoin went live alongside them. All of it was pulled with Bitquery's trade API.
| The record | |
|---|---|
| First pump.fun trade priced in a stock | 9 September 2026, 19:40:17 UTC, in SPYx |
| SP500 xStock (SPYx), Solana | XsoC…DF2W |
| NVIDIA xStock (NVDAx), Solana | Xsc9…9qEh |
| Anthropic PreStocks, Solana | Pren…Lkhw |
| Invesqo QQQ (QQQB), BNB Chain | 0x2058…efc7 |
| SpaceX bStock (SPCXB), BNB Chain | 0xbe9d…03e1 |
| Cea Industries 4Stock (BNC4), BNB Chain | 0x7c8d…545c |
| Busiest seller-without-buying wallet | appears in 9,647 separate coins |
Ask these questions in plain English
Every figure above came from queries anyone can run. The Bitquery MCP server puts the same index behind an AI assistant, so you can ask which coins are priced in a given stock token, when a launchpad started offering it, what a wallet paid in and took out, or how a new coin's first ten seconds went, without writing the query yourself.