Buybacks: Do They Actually Work?
Who is actually creating value? Buybacks are easy to announce. This report tests what funds them, what happens to the tokens, and whether the program can survive after incentives, costs, and dilution.
- Coverage
- Token Buybacks
- Evidence cutoff
- July 29 · 19:25 UTC
Chapters
A buyback only creates value when real earnings fund a real reduction in supply.
- Buybacks are not earnings. Fees are what users pay. Earnings are what remains after rewards, incentives, operations, bad debt, taxes, and other claims on the same cash.
- HYPE and PUMP lead the current revenue group. DefiLlama recorded $37.5 million and $20.3 million of 30-day protocol revenue, respectively. That measures scale, not audited profit.
- The token destination changes the result. HYPE and PUMP are burned. dYdX is staked. Pendle distributes purchases to active sPENDLE holders. Raydium and Jupiter hold purchased tokens, which can return to circulation later.
- CAKE shows what a complete supply bridge looks like. PancakeSwap reported 652,564 CAKE minted and 2.40 million burned in June, producing a net reduction of 1.75 million CAKE. It has now reported 34 consecutive months of net supply reduction.
- Aave proves that continuity matters. The DAO spent $42 million to acquire more than 205,000 AAVE, then paused the program after a bridge incident to protect its balance sheet.
- Clear rules are valuable even at smaller scale. Pendle directs 80% of eligible fees to purchases. dYdX directs 75% of net protocol fees to monthly buy-and-stake operations.
- Traders need a cash-flow and dilution bridge. The useful comparison is recurring external revenue, minus costs and token outflows, followed by visible execution. An announcement dollar is not enough.
The headline says buyback. The balance sheet decides whether it matters.
A protocol can buy its token with customer revenue, treasury reserves, newly raised capital, or money that should have covered another obligation. Those programs may look similar on a dashboard, but they do not create the same value.
Start with the cash. Outside demand means a customer paid for trading, lending, blockspace, or another service. Gross fees may then be split among liquidity providers, validators, affiliates, or token incentives. Only the amount left after those claims can begin to resemble operating surplus.
Then follow the token. A permanent burn reduces total supply. A staked or treasury-held token may support the network, but it still exists. A distributed token transfers value to one group of holders and may be sold back into the market.
Ask whether repeat customer cash is buying tokens faster than the protocol is creating, unlocking, incentivizing, or redistributing them.
The current market has two scale leaders and a long tail of smaller programs.
Hyperliquid and Pump.fun generated the largest 30-day protocol revenue in the selected group. That gives them more room to fund purchases, but scale alone does not prove profit or a fair valuation.
DefiLlama reported $37.5 million of 30-day protocol revenue for Hyperliquid and $20.3 million for Pump.fun at the cutoff. Sky followed at $13.7 million. Every remaining program was below $4.2 million.
The comparison has a hard limit. Each protocol reports a different business. Trading fees, lending spreads, liquidation income, swap fees, and launchpad revenue do not have identical cost structures. The chart answers who has the largest reported revenue base. It does not answer who has the highest net margin.
“Buyback” describes at least four different token outcomes.
The allocation rate tells only part of the story. Investors also need the funding base, execution rule, destination, and current operating state.
Hyperliquid converts part of its fee flow into HYPE inside the L1, then burns the tokens held by the assistance fund. Pump.fun states that half of every dollar it earns is used to buy and burn PUMP. Both create a clear link between protocol activity and permanent token removal.
Pendle and dYdX use a different model. Pendle buys PENDLE and distributes up to all of it to active sPENDLE holders. dYdX buys and stakes DYDX. The economic benefit depends on who receives the tokens, how long they remain locked, and whether they can return to the market.
Raydium uses 12% of trading fees to buy RAY, then holds the tokens at a public address. Jupiter's reported 50% allocation also accumulates tokens rather than burning them. Treasury accumulation can still matter, but it is not permanent supply removal.
Large revenue supports the story, but the path has been volatile.
HYPE has the strongest absolute revenue base in the cohort. PUMP has a smaller business and a much smaller market value, which makes its buy-and-burn flow look larger relative to the token.
Hyperliquid generated $782.5 million of protocol revenue over the trailing year in the provider series. The monthly path peaked above $113 million in August 2025, fell sharply into 2026, and recovered to nearly $60 million in June. July is partial through the evidence cutoff.
Pump.fun generated $325.4 million over the trailing year. Its first-party page showed $416 million of cumulative buybacks and 154.43 billion PUMP burned at the cutoff. That cumulative number is larger than the current trailing-year revenue because it spans a longer program history.
Visible purchases matter more than a governance promise.
A program becomes credible when investors can verify the cadence, the amount spent, the tokens acquired, and the final destination.
Pump.fun provides the clearest large first-party counter. Its token page displays dollars spent and tokens burned. dYdX publishes a buyback dashboard and reported 8.46 million DYDX bought and staked as of January 1, 2026.
Aave also disclosed strong historical execution. The DAO used $42 million to acquire more than 205,000 AAVE during the first ten months of its program. The same record shows why a purchase total is not the whole thesis. Aave later paused purchases after an rsETH bridge incident so the DAO could preserve its balance sheet.
The missing proof is a bridge from cash to surplus and from buybacks to net supply.
No program in the main comparison publishes one complete statement that starts with recurring customer revenue, subtracts every operating and balance-sheet claim, then reconciles the resulting purchases against all token issuance.
PancakeSwap provides the strongest net-dilution example. Its June report lists 652,564 CAKE minted, 2.40 million burned, and a net supply reduction of 1.75 million CAKE. Cumulative net reduction reached 52.80 million CAKE across 34 consecutive months. That does not prove audited free cash flow, but it proves the final supply result.
Spark provides the stronger cash-flow bridge. Its official data hub showed projected yearly net returns of $27.70 million, operating expense of $14.40 million, and projected protocol surplus of $13.30 million. It also showed 72.18 million SPK purchased at an average price near $0.0228. The remaining problem is dilution: Spark separately discloses a ten-billion-token emission schedule over ten years, so purchases cannot be treated as net deflation by themselves.
The best disclosed programs make every step easy to follow.
High-quality buybacks connect a recurring source of outside cash to a visible purchase rule and a clear token destination. They also disclose the liabilities that compete for the same cash.
HYPE and PUMP score well on revenue scale, execution visibility, and permanent burn. Their remaining investor question is not whether tokens are being removed. It is whether future revenue can outrun changing market activity, unlocks, incentives, and valuation.
Pendle and dYdX score well on rule clarity. Their smaller revenue base makes consistency more important. Raydium is transparent about its percentage and address, but holding RAY does not reduce total supply. Aave's pause lowers current continuity even though the historical record is unusually detailed. CAKE and SPK are accounting benchmarks rather than direct additions to the score: each proves one side of the bridge that the main programs still leave incomplete.
The market will trade the ratio before accounting catches up.
Annualized 30-day protocol revenue divided by market value can identify where a small token sits beside a large reported revenue stream. It cannot be called price-to-earnings.
PUMP screens at 32.3% on this gross revenue proxy. SKY and GMX screen above 11%. Those numbers can look cheap, but they ignore operating costs, incentives, bad debt, treasury obligations, unlocks, and future issuance.
The trade works when three forces align: revenue grows, the allocation remains active, and net token supply tightens. It weakens when volume cools, governance reduces or pauses the program, or new token supply overwhelms purchases.
A durable program should get easier to verify every month.
The next proof is not another announcement. It is a monthly bridge from customer revenue to cash available for purchases, followed by the exact change in token supply.
Separate recurring user fees from liquidation spikes, one-time launch income, treasury yield, and accounting reclassifications.
Track whether operations, incentives, bad debt, taxes, and reserves are covered before purchases occur.
Verify purchase dates, dollars spent, token amounts, average prices, destination addresses, and any pauses.
Compare burned or locked tokens with emissions, unlocks, incentives, treasury transfers, and returned tokens.
Watch for budget cuts, parameter changes, emergency pauses, and new claims on the same cash flow.
Compare price with revenue, buyback cadence, liquidity, depth, and supply change. Price alone cannot confirm the thesis.
The data can rank scale and disclosure. It cannot yet rank true net earnings.
This report freezes market and revenue data at one cutoff, then uses first-party documents to classify how each program works.
Methodology and limitations
Thirty-day and one-year revenue values come from DefiLlama's dailyRevenue series. Definitions vary by protocol and do not equal audited net income. The report uses the series to compare scale, not margins.
Market values and prices come from a fixed CoinGecko snapshot. The gross revenue proxy annualizes the latest 30 days and divides that amount by market capitalization. It ignores operating costs, incentives, taxes, bad debt, treasury needs, unlocks, emissions, and other token outflows. It is not price-to-earnings and not earnings yield.
Allocation, status, execution, and token destination come from official documentation, governance, or first-party dashboards. JUP and SKY parameter status is marked medium confidence because the current configuration is not presented in one complete authoritative record.
Cumulative execution snapshots use different dates and program histories. They prove that purchases occurred. They should not be used as a direct performance ranking.
CAKE and SPK are accounting comparators. CAKE's monthly burn report reconciles token mints and burns but is not a complete post-expense cash statement. Spark's financials bridge projected returns to operating expense and surplus, but its buyback must still be compared with the separate long-term emission schedule.
Grok Heavy and GPT Pro were used only to find possible omissions and contradictions. No AI output is cited. Every published claim is tied to a direct protocol, governance, onchain, or named market-data source.
This report is for information and research. It is not investment advice or an offer to buy or sell an asset. Digital assets can lose substantial value. Verify current program rules, treasury health, token supply, contracts, and market data independently.