Launches are back, discovery isn't

01Put both curves on the same ruler first

The most common wrong comparison I've seen in recent weeks puts "primary is still 43% below its peak" and "primary is up 192% from the bottom" in the same paragraph. One is measured from the top, one from the bottom; readers can't compare them. This report uses a single ruler throughout: what percentage of its own all-time peak. The chart below adds one more layer: bars are monthly fees in dollars, lines are each track's share of that month's combined total.

Absolute amounts make it clearer. In January 2025 the secondary discovery layer collected $302.6M in fees in a month, 1.9× the primary market ($161.4M) — back then the money sat mostly at the "finding coins" end. By August 2026, primary was $91.25M and secondary $86.68M: for the first time, secondary is smaller than primary.

Definitions: primary = sum of the top 60 in DefiLlama's "Launchpad" category. Secondary discovery layer = the line-up of trading terminals and bots — Axiom, GMGN, Photon, BullX, Trojan, Maestro, BONKbot, DEX Screener, Banana Gun, fomo Wallet, tradeXYZ, Terminal, pump.fun mobile and others — excluding the pump.fun main site and Jupiter to avoid double counting with primary.

02Reason one: what recovered is not trading, but the act of launching itself

This is the easiest one to skip and the one that explains the most. Primary-market money and secondary-market money don't come from the same thing at all.

• Launchpad revenue = launch fees + trading fees (or tax) on the curve, largely unrelated to whether anyone trades the coin later;
• Terminal revenue = user volume × take rate; someone has to actually trade in the secondary market for it to earn anything.

Most of primary's growth in August 2026 came from Flap on BSC, and Flap's growth is pure throughput: launches jumped from 238K in July to 1.318M in August (5.5×), while fees per coin only moved from $7.55 to $7.90. Over a longer window, monetization per coin fell 57% in eight months ($18.38 in January 2026).

Context that must come with it: a graduation rate around 2% is normal for these platforms (pools.trade 2.43%, Pons 2.37%). Flap's 0.24% is an order of magnitude lower not because the coins are worse, but because the denominator is machine-made — a launch costs only $0.69 with no review at all. Pons launched 1,895 coins in total; Flap launches 70K in a day.

03Reason two: issuers swallowed the discovery step themselves

In 2024 the division of labour was: pump.fun launches → Raydium provides the pool → terminals like Photon / BullX / Trojan handle discovery and orders, each collecting its own fee. By 2026 issuers had taken back every link in that chain.

Tougher still is pump.fun's new fee schedule effective 1 September 2026: after graduation, the total fee on the PumpSwap canonical pool falls monotonically with market cap from 1.25% to 0.30%, so bigger coins pay less. The only purpose of this pricing is to keep big coins in its own pools and away from outside pools and aggregators. The better issuers get at keeping users, the less toll the discovery layer collects.

04Reason three: not a general shrink, but a generation replaced

Treating "secondary discovery layer −71%" as a smooth downhill is wrong. Look at each venue's retention from its own peak to now and you see a complete generational turnover.

The result is rising concentration: the top two's combined share of the secondary discovery layer went from 46.0% in January 2025 to 62.2% in August 2026. This is not a market with room left; it has just been reshuffled and the winners are already in place. Axiom 34.3% + GMGN 27.9%: 62% combined.

05Reason four: growth moved to chains the terminals don't cover

That terminal line-up is almost entirely Solana-native. Primary-market growth in 2026 is mostly not on Solana.

The last column, "30d / total", tells you how new a chain's business is:

• Solana is only 3.2% — of $1.57B cumulative, the last 30 days are one-thirtieth. It's an old chain; its money was earned over the past two years.
• Robinhood Chain 35.3%, X Layer 49.2%, Monad 12.6% — half their business happened in the last month.
• BSC at 3.1% also looks "old", but that's diluted by four.meme's historical total; BSC's current growth is concentrated in Flap alone.

Robinhood Chain deserves a separate note: it reached the sector's second-largest 30-day volume with 46 launchpads (more than Solana's 43). Many launchpads is itself a sign of traction — a chain with only three to five launchpads usually means no one wants to come, not that the few left are strong. That's also why you can't rank chains by "average fees per launchpad": fewer shops mechanically means a higher average; it only reflects concentration.

The implication for the discovery layer is direct: this new volume happens on Robinhood Chain, X Layer, Monad and BSC, while the 2024–2025 Solana terminals mostly don't support these chains at all. They weren't beaten; they're standing at an intersection with no foot traffic.

06Reason five: the tax-token model redistributed "fees"

Flap, the main driver of BSC growth, uses a tax-token model, which differs fundamentally from the take-rate model. I read the contracts end to end:

The token contract FlapTaxTokenV3 doesn't distribute money itself; once enough tax accumulates it calls each token's own TaxProcessor clone. The split is hard-coded in TaxProcessorBase.sol: first the platform's share is taken at feeRate, then the rest is split into four buckets — project wallet / burn / liquidity / holder dividends — and the contract enforces that the four sum to 10000.

So 90% of the tax users pay goes back to holders and burns; only 10% goes to the platform. This model keeps "the money trading generates" inside the token instead of handing it to outside terminals and aggregators. A terminal that only forwards orders has nowhere to stand on this path.

Measured on-chain, calling each token's feeConfig(): token implementation 0x024f18294970B5c76c0691b87f138A0317156422, TaxProcessor implementation 0x9fe67a78f05c6569db77e3f4d957ed67bddc9569, Dividend implementation 0x666c0a0e0cccb2305bc2d9aa6445c3ff076bb031. Each token is its own EIP-1167 clone with its own configuration.

07Ranking the five reasons

08If you're building a discovery product, where does this data point

Read the five reasons backwards and you get what this market is missing now:

• The bottleneck has moved from "can you see it" to "is it worth seeing". Building a terminal in 2024 solved latency, and that problem is solved — which is exactly why pure-speed products all went to zero. Today's problem is how to find the 15 coins among 70K new ones a day.
• Filter the denominator, not the coins. Survival of 0.24% vs 2.4% is an order of magnitude apart, and the gap isn't coin quality but launch cost ($0.69 vs real money). Any ranking that doesn't first cut the denominator drowns in machine-made noise.
• Cross-chain is a hard requirement, not a bonus. Growth is on Robinhood Chain (30d / total 35.3%), X Layer (49.2%) and Monad (12.6%). Solana-only means serving a chain that has already earned its money.
• Don't count on monetizing via order take rates. Issuers are systematically keeping trading at home — pump.fun with lower fees for big coins, Flap with tax dividends inside the token. The ceiling on take rates is falling.
• The two survivors show the shape of the answer: GMGN sells address labels and copy-trading (information); Axiom sells an all-in-one front end and rebates (distribution). Neither is "a faster button".

09Method and limits

• All fee figures come from DefiLlama Fees API dailyFees, aggregated by UTC month, as of 2026-08-29; that day is incomplete and excluded.
• Primary = sum of the top 60 in the Launchpad category; 81 tail venues are excluded, so the true total is slightly above the figures here.
• Secondary discovery layer = trading terminals and bots, excluding the pump.fun main site and Jupiter. The list is hand-picked; a different list changes the magnitude, not the direction.
• By chain, multi-chain protocols are divided by their number of chains, so chains add up to the sector total; booking the full amount on every chain would raise Solana's and Robinhood Chain's absolute values without changing the order.
• Survival, launch counts and split ratios come from BSC public nodes and verified source code and are reproducible.