On the bonding curve, one wallet made almost a third of the trades.
A Pump.fun chart full of green and red candles looks like a crowd. We read five and a half minutes of Solana blocks and grouped every Pump.fun transaction by the wallet that signed it. The crowd turned out to be a lot smaller than the charts suggest.
How we counted
We read 1,550 consecutive Solana blocks, slots 471,641,813 to 471,643,362, produced between 06:46:17 and 06:51:51 UTC on November 20, 2026. Those blocks held 438,577 successful transactions. We kept every one that touched either the Pump.fun bonding curve program or PumpSwap, the exchange where coins trade once they graduate. That left 105,696 transactions.
For each one we wrote down the signer, which is the wallet that signs the transaction and pays its fee, plus the fee, any Jito tip, and every token balance that changed. Failed transactions are not in the sample.
One warning before the numbers. A wallet is not a person. One trader can run fifty wallets, and one trading app can sign on behalf of thousands of users. When we say “wallet” below, we mean exactly that and nothing more.
A quarter of the block space
24.1% of all successful transactions in those blocks touched Pump.fun or PumpSwap. The median block carried 63 of them and the busiest carried 324. Whatever you think of memecoins, they are a big share of what Solana validators process on an ordinary weekday morning.
Most of that traffic was not on the bonding curve. 79.3% of the transactions went to PumpSwap and only 20.7% to the curve. That lines up with what we found in an earlier piece here, when we tracked 1,634 coins after graduation: the 31 coins that had graduated within the hour carried more than 64,000 trades between them. Graduated pools are where the volume sits.
16,753 wallets, but a few did most of the work
The 105,696 transactions came from 16,753 signing wallets. More than half of them, 9,728 wallets or 58.1%, signed exactly one transaction in the window. Together those one-time wallets account for just 9.2% of the traffic.
At the other end, 132 wallets signed 50 or more transactions each. That is 0.8% of the wallets and 44.7% of all transactions. The ten busiest wallets signed 25,516 between them, close to a quarter of everything. The single busiest signed 5,522, which is more than three per block for the whole window. Nobody clicks that fast. These are programs.
1 transaction: 9,728 wallets (58.1%), 9.2% of all transactions2 to 9 transactions: 5,604 wallets (33.5%), 20.8%10 to 49 transactions: 1,289 wallets (7.7%), 25.3%50 or more transactions: 132 wallets (0.8%), 44.7%
That shape is not unusual for a busy market. A small group of automated traders does most of the clicking, and a long tail of people show up once. What is unusual is how far the gap stretches in some corners of Pump.fun.
The bonding curve is a small room
Split the sample by venue and the picture changes a lot. For this part we only counted transactions that actually moved tokens. The ones that did not get their own section below.
On PumpSwap, 57,472 token-moving transactions came from 15,240 wallets, and the ten busiest made just 4.3% of them. That is what a broad market looks like.
On the bonding curve, 17,129 token-moving transactions came from only 1,874 wallets. One wallet made 31.9% of them. The top three made 49.3% and the top ten 55.4%. The busiest one traded 109 different coins in five and a half minutes, almost all of it on the curve, and paid a 45,000 lamport fee on a typical transaction.
Chain data alone can’t tell us what that wallet is. It could be a sniping service buying for many users, a market making program, or one trader with a fast script. What the number does tell you is practical. On a coin that is still on the bonding curve, a lively trade feed can come mostly from one or two programs. Before you read activity as demand, open the trade list and look at how many different wallets are behind it.
Almost a third of the transactions moved nothing
31,095 transactions, 29.4% of the sample, succeeded, touched a Pump.fun or PumpSwap program, and changed no token balance at all. The signer paid a fee and nothing was bought or sold.
They came from just 177 wallets. For wallets that signed 50 or more transactions, 64.6% of what they sent was of this kind. For one-time wallets it was 0.4%. A single program that we have not identified sat inside 15,826 of them.
The most likely explanation is arbitrage and similar trading programs. They send a transaction that checks for a price gap and exits quietly when the gap isn’t there. On Solana that check is cheap enough to repeat thousands of times: all 31,095 of these transactions together cost their signers about 0.23 SOL in fees and tips.
This matters if you use transaction counts as a sign of interest in a coin. Some of what shows up as activity on a pool never bought or sold a single token.
The five busiest wallets, side by side
Looking at the top five signers one by one shows how different “busy” can be. Only two of them were mostly trading.
Wallet 1: 5,522 transactions, 5,470 on the bonding curve, 109 different coins, 1 that moved nothingWallet 2: 4,455 transactions, all on PumpSwap, 4 coins, 4,448 that moved nothingWallet 3: 3,147 transactions, all on the bonding curve, 1 coin, 3,145 that moved nothingWallet 4: 2,757 transactions, almost all on the bonding curve, 20 coins, 450 that moved nothingWallet 5: 2,022 transactions, all on PumpSwap, 3 coins, 2,013 that moved nothing
Wallet 3 is the odd one. It sent more than 3,000 transactions at a single coin on the curve in five and a half minutes, and almost none of them changed a balance. If you were watching that coin in an explorer that lists every transaction, it would have looked like the busiest coin on Pump.fun. Nothing was changing hands.
What one transaction costs
Every Solana transaction pays a base fee of 5,000 lamports per signature and can add a priority fee to move up the line, as the Solana fee docs explain. One lamport is a billionth of a SOL.
The typical Pump.fun transaction in our sample paid 5,369 lamports in total fees, barely above the base. 41.2% paid the base fee and nothing more. The spread is wide, though. The 90th percentile paid 45,000 lamports and the 99th paid 704,999. The most expensive 1% of transactions paid 62.5% of all the fees and tips in the sample.
Jito tips, the extra payment that sends a transaction through a Jito bundle, were rare. 3,066 transactions, 2.9% of the sample, carried one, and the median tip was 2,744 lamports. Busy wallets tipped more often: 4.2% of transactions from wallets with 50 or more carried a tip, against 0.8% from one-time wallets.
One-time wallets paid a median of exactly 5,000 lamports. People who make a trade now and then are mostly not in a fee race. The race is between programs.
Small buys pay more, as a share
A flat fee weighs more on a small trade. We took 4,775 bonding curve buys where we could read the SOL spent and compared the network cost, fee plus any tip, with the size of the buy. A quarter of curve buys were under 0.01 SOL, and the median buy was about 0.06 SOL.
Under 0.01 SOL (1,169 buys): 0.80% of the buy went to network costs0.01 to 0.05 SOL (839 buys): 0.065%0.05 to 0.1 SOL (634 buys): 0.033%0.1 to 0.5 SOL (1,470 buys): 0.025%0.5 to 1 SOL (344 buys): 0.017%1 SOL and up (319 buys): 0.007%
These figures cover the Solana network only. They leave out the trading fee the venue charges and the price impact of the trade itself. For most buys the network cost is a rounding error. Under 0.01 SOL it starts to show, and a high priority fee preset in a trading app can multiply it several times over.
What to take from this
Count wallets, not trades. A coin with 400 trades from 12 wallets is a very different thing from 400 trades from 300 wallets.Be extra careful on the bonding curve. With under 2,000 active wallets across the whole curve in our window, one program can make a coin look busy.Transaction counts include checks that bought nothing. If a dashboard counts every transaction that touches a pool, part of the number is arbitrage traffic.Look at your own priority fee setting. On a small buy, a fee preset built for racing can cost more than it saves.
Limits of this sample
This is five and a half minutes on one weekday morning, so the exact shares will move from hour to hour. We grouped by signer, and in 10,671 transactions the tokens moved for a wallet other than the one that signed, which is what you see when an app or relayer signs for a user. Those are counted under the signer.
We left failed transactions out. On busy minutes failed transactions can be a large share of attempts, so the real number of tries is higher than what we counted. We also did not try to name the programs behind the busiest wallets. Everything here comes straight from Solana blocks read through a standard RPC node, with no third-party labels on top.
Who Really Trades on Pump.fun? We Sorted 105,696 Transactions was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.
