Raydium has run a systematic fee-funded buyback since early 2024. The engine works — on-chain, every day. The question is who ends up owning its output. RAY is a claim on ~11 cents of every fee dollar, delivered as open-market purchases that accumulate in a protocol-controlled wallet now holding more than 30% of circulating supply: a genuine demand sink, and a treasury asset holders have no enforceable right to. September is why the token moved. Quarterly fees had fallen 96% — from ~$412M in Q4 2024 to ~$17M in Q2 2026 — after Pump.fun took its graduations in-house; then StonkFun and tokenized equities pushed 30-day fees back to ~$41.7M and RAY rose ~150% in thirty days. At ~$543M the token trades at ~30× trailing holder revenue but ~10× the September run-rate, which means the market is underwriting roughly half of September’s pace as durable — on a niche whose distribution Raydium does not own. The plumbing is built and it has survived a legacy-code exploit and the loss of its largest customer without minting a single token. What is missing is the last step: turning an accumulating treasury into retired supply and a disclosed cap table.
Serum's arm, then Pump's pipe
Raydium was born as infrastructure for someone else’s exchange, and its history is a sequence of being the default venue for flows it did not originate — then losing them when the originator integrated vertically. It launched in February 2021 as a hybrid AMM on Serum with a fixed 555M supply, hit its all-time high of ~$16.8 that September, and survived both FTX’s collapse and a December 2022 key compromise that drained ~$4.4M. Buybacks went systematic in early 2024, at 12% of fees, in the same window that team vesting ended on 21 February. By Q1 2025 it was taking ~50% of Solana DEX volume on ~$407M of quarterly fees. Then on 20 March 2025 Pump.fun launched PumpSwap and stopped migrating graduations, and the fee line began the fall that defines the rest of the story.
On insiders the picture is unusually clean in one direction and unresolved in another. There is no disclosed institutional VC round, no cliff left, the mint authority is disabled and the buyback flow is auditable; the team shipped for five years and repaid LPs after both exploits. Against that: the founders are pseudonymous, the foundation is Cayman-based and nominee-directed, and ~285M RAY — about 51% of max supply, mostly the mining reserve (34%, emitting ~1.9M a year) and partnership and ecosystem (30%) — sits outside circulation, published by bucket with on-chain addresses but with balances this desk has not reconciled against them. The same small multisig holds the >81M-RAY buyback stack. The report’s own status note says it plainly, and so does this page: the team is aligned with the protocol; alignment with holders is partial, because nothing binds how the dormant supply or the bought-back stack gets used.
A fee line that can fall 96%
Raydium runs Solana’s broadest liquidity toolkit — AMM v4, CPMM, CLMM, LaunchLab, perps — and keeps shipping. None of it requires RAY, and every program remains upgradeable by a 3-of-4 multisig behind a 24-hour timelock. The numbers are fully on-chain, and what they show is amplitude: gross quarterly fees ran 79, 72, 102 and 412 through 2024, 407 in Q1 2025, then 86, 114, 48, 35 and 17 as PumpSwap took the graduations, before Q3 2026 came back to ~54 by 28 September. Thirty-day fees of $41.7M annualise to roughly $500M against $154.9M actually booked over the trailing year — which is the single most important reason to discount the September number rather than extrapolate it. The surge is concentrated in StonkFun and tokenized-stock listings, and StonkFun was integrated on 5 September.
The competitive map explains why the collapse happened and why it can happen again. Solana’s DEX market reorganised around vertical integration: the launchpad that owns the user now owns the AMM too, which is why PumpSwap took ~67% of indexed volume by 23 September 2026 while Raydium’s share went from ~50.3% in Q1 2025 to ~5.1%. Even Uniswap entered Solana as a Jupiter-routed interface rather than a liquidity venue, because routing — not pooling — is where the leverage sits. Each rival also routes value to its own token differently: PumpSwap burns PUMP with 50% of net fees, Jupiter runs JUP buybacks, and Raydium points 12% of fees at RAY and holds it. Raydium’s answer is to be the default for what the integrators do not run themselves — tokenized stocks and partner launchpads, with LaunchLab turning third-party launchpads into fee sources. It works, with one caveat the numbers make hard to ignore: on LaunchLab in Q3 to date the platforms kept ~4× Raydium’s own protocol fee ($6.66M against $1.63M), and ~6.6× in the week to 12 September.
Bought back is not burned — what a dollar of buyback costs
Follow a Raydium fee dollar and the split is not close: across Q3 2026 to date, of $53.75M gross, $44.9M (83.6%) went to LPs, launch platforms and creators, $5.9M (11.0%) to RAY buybacks and $2.9M (5.5%) to the treasury. That is the book’s thesis in one chart — on trading venues most fee value reaches the LP and whoever owns distribution, not the governance token — and RAY’s 12% is a policy, set by a 3-of-5 treasury multisig with no timelock, rather than a right a holder can enforce. The disposition problem sits one layer below: AMM buybacks accumulate at a public address controlled by the protocol multisig with no disclosed constraint, and DefiLlama describes LaunchLab’s 25% slice as buy-and-burn with only a small burned share, not yet confirmed on-chain. The one hard guarantee in the whole structure is the disabled mint — nobody controls it, and it is immutable.
Four moments tell you what this protocol is. December 2022: a pool-authority key was compromised for ~$4.4M, and buyback funds served the LPs, not the token. March 2025: Pump.fun left and quarterly fees went from ~$407M to ~$86M — the partner that owns the user takes the fee pool with it. June 2026: a legacy AMM V3 exploit cost ~$1.34M, treasury-refunded, because dead code still carries live risk. September 2026: the StonkFun surge, on a partner integrated that month — concentration is back, in a new costume. Holder economics are real; holder control is nil. So the honest answer to whether RAY wins when Raydium wins is: partially, at a fixed ~11% of the fee line, delivered into a wallet the holder does not control. Better than most DEX tokens, worse than the headline “30% bought back” suggests. Since early 2024 Raydium has pointed a fixed slice of real fees at its token through an exploit and the loss of its largest customer without minting a single token; what is missing is the last step — turning an accumulating treasury into retired supply and a disclosed cap table. The plumbing is built and the covenant is pending, and for a patient holder that gap is exactly where the opportunity sits.
Key findings
- The accrual is real, systematic and auditable. RAY receives ~11 cents of every fee dollar — 10.6% over thirty days ($4.43M of $41.68M) and 11.0% in Q3 2026 to date — as open-market purchases running daily since early 2024, with $190.4M of buybacks (69.1M RAY) recorded as of July 2025 at a public address. The mint authority is disabled, emissions run at ~1.9M RAY a year against 3–10% of float absorbed annually, and there is no institutional VC round disclosed and no cliff left. On supply mechanics this is cleaner than most DEX tokens.
- But bought back is not burned — and that is the whole argument. The absorbed stack, more than 30% of circulating supply (≈81M+ RAY), sits in a wallet controlled by the protocol multisig with no disclosed constraint: it can be burned, locked, sold or spent. That is not hypothetical — in 2022 buyback funds were used to repay exploit victims, which served LPs rather than the token. RAY carries no fee-share, no fee yield and no binding vote. The held stack is a cushion for the protocol, not a claim for the holder.
- The fee line fell 96%, then tripled in a month. Gross quarterly fees went from ~$412M in Q4 2024 to ~$17M in Q2 2026 after Pump.fun launched PumpSwap on 20 March 2025 and stopped migrating graduations to Raydium. September 2026 reversed part of it: StonkFun and tokenized-stock listings pushed thirty-day fees to ~$41.7M (≈$500M annualised against $154.9M trailing twelve months), thirty-day holder revenue to ~$4.4M, and TVL to $1.33B, up 18.8% in thirty days. RAY rose ~150%. Amplitude, in both directions, is the defining feature of this business.
- Distribution is rented, and the share already collapsed once. Raydium held ~50.3% of Solana DEX volume in Q1 2025 and ~5.1% by 22 September 2026, while PumpSwap held ~67% of indexed volume — the market reorganised around vertical integration, where the launchpad that owns the user owns the AMM too. The new niche is real (63% of Solana tokenized-stock DEX volume per Messari, >90% per Raydium, ~$2.3B in Q3) but it is the same shape as the old one: a default venue for flows it did not originate. The Pump.fun precedent is the template, and StonkFun was integrated on 5 September.
- What the price already assumes. At a ~$543M market cap, RAY trades at ~30× trailing holder revenue (~$18.0M) and ~10× the September run-rate (~$53M); on protocol revenue it is ~22× trailing, and on FDV ~$1.12B the same lenses read ~62× and ~21×. At an illustrative ~20×, $543M implies ~$27M a year of durable buybacks — about half of September’s pace. Spot at ~$2.01 sits inside the base case of $1.65–2.85. The asymmetry is in the disposition: burn the stack and the denominator falls from 269.7M to ~189M, lifting the bull range from $4.45–7.40 to ~$6.35–10.60.