Strip out the two names and the record stops looking like an adoption curve. Nearly 90% of the $638M that the Financial Times, working from Allium Labs data, attributes to token buybacks belongs to Hyperliquid and Pump.fun [16], which leaves roughly $64M for every base layer chain and blue-chip DeFi treasury that has since taken up the mechanic [1]. The concentration is also not new: in August 2025 weekly repurchases ran near $40M, with Hyperliquid at about $24M and Pump at about $10M [5], or 85% of the week between them [9].
Lex Sokolin of Generative Ventures put the scaling problem plainly, saying a $100M annual program is irrelevant against $20M of daily trading [7], which is about $274,000 a day of buying against $20M of turnover, call it 1.4% of the flow [3]. Amir Hajian of Keyrock adds the other side of the ledger, where unlocks and constant emissions undermine whatever the buyback bids for [8]. That ratio, not the treasury policy, decides whether a buyback is a bid or a press release.
So the aggregate is the wrong instrument. Dollar-weighted, the trade worked: HYPE is up 145% and PUMP 109% year to date while bitcoin fell 10% and total crypto market capitalisation fell 11.9% [18], a spread of roughly 157 points between HYPE and the market it trades in [5]. Equal-weighted it did not, and the tokens that stayed flat or fell through 2025 despite active programs [3] are the ones whose revenue is small relative to their float turnover.
This is probably wrong, but the mechanism I would bet on is that buyback spend works as the cleanest public proxy for revenue, and revenue is what got repriced. Hyperliquid runs about 99% of more than $800M of annual revenue through buy-and-burn and has retired some $1.3bn of HYPE since November 2024 [13]; its own second-quarter print, $169M of revenue and $141M directed at HYPE [19], is 83% [2], or rather the more interesting version, a payout ratio that moves with the quarter rather than a constant. Pump.fun's half of $420M annualised revenue is nearer $210M a year [20][6]. Bitwise's Matt Hougan said in August that valuations could double in two years as protocols route revenue this way [21], which is the same claim with the causation running the other direction.
Hyperliquid funds nothing else with that revenue, because 99% leaves no room for an incentives budget or a treasury buffer. And the structural break with equities is that most token holders have no legal claim on protocol cash flow, so retiring float does not concentrate ownership the way a share repurchase does, per Hajian [9]. Tokenomist's wider count of $18.8bn across 27 tokens from January 2025 to July 2026, of which buybacks were only about $2.8bn [6], puts purchases at roughly 15% of the story [4]; of 11 programs with clean on-chain data, Jupiter parks repurchases in a trust and Aster hands them to stakers, so neither shrinks supply [11], and among newer emission tokens only BNB and RAY actually did [12].
What would prove this wrong is cheap to specify. If HYPE and PUMP roll over while revenue and buyback spend hold, the scale story survives and my revenue-proxy read dies. And if the $638M really is year-to-date through Aug 25 [17], it annualises near $986M against $1.4bn of buybacks in 2025 [2][8], and the spending is shrinking.
