Hyperliquid's Flywheel, Measured: Fees, Buybacks and the Second Engine

Revised Sep 14, 2026: this report has been rebuilt on measured data — the Hyperliquid API fee series, the weekly fee-to-buyback reconciliation, and on-chain unlock records. Figures that cannot be verified are explicitly marked unavailable or Modeled, including the HyperEVM segment split we previously published in error.

The Hyperliquid investment case is a flywheel story: trading generates fees, fees fund buybacks, buybacks tighten supply, and the cycle compounds. Stories like that are easy to assert and hard to audit. This report audits it — number by number, from the API up. The findings: the fee engine is real and just had its best month ever ($67.3M in August); the conversion machinery reliably turns roughly 97% of net fees into HYPE buybacks; and a second engine is being bolted on — AQAv2's reserve-yield stream — whose first verifiable print arrives Oct 3. Also findings: the parts of the flywheel that get the most airtime, like the HyperEVM DeFi fee base, are the parts we currently cannot measure — and we say so plainly rather than estimate them.

1. Background — the two engines and the pipes between them

Hyperliquid's revenue architecture has exactly two taps and one drain. Tap one: trading fees — taker and maker fees across core perps, spot and builder markets. Tap two (new): reserve yield — as of Aug 26, the AQAv2 upgrade routes roughly 90% of the yield on the platform's USDC reserves to the Assistance Fund, with Coinbase managing the balances on Circle's infrastructure. The drain: supply — scheduled unlocks that add circulating HYPE, which buybacks must absorb for the flywheel to tighten anything.

The pipes are the part that makes Hyperliquid unusual: essentially all net trading fees (~97%) are programmed into HYPE buybacks via the Assistance Fund. Most venues retain revenue; Hyperliquid converts most of it into mechanical token demand. Whether that conversion is meaningful is an empirical question with three measurable parts: how big the fee tap runs, how much survives the drain, and whether the new tap adds anything. Sections 2–5 measure each.

2. The fee engine, measured

Pulled from the protocol's own fee endpoint, week by week — no estimates:

Three honest readings. Level: August totaled $67.3M — annualized, a ~$0.81B run-rate, which would rank the venue's fee generation among the top protocols in crypto by most published league tables. Trend: the engine is rate-sensitive, not constant — three consecutive weekly declines ($24.6M → $21.2M → $16.5M → $12.9M) followed the price fade, and the trailing run-rate is nearer ~$0.67B annualized than August's peak. Floor: even the weakest recent week ($12.9M) sits well above the pre-August quiet baseline ($8–9.4M). The engine moved up a gear in August; September is testing whether the gear holds.

3. From fees to buybacks — the conversion machinery

This is the pipe, and it reconciles cleanly against published buyback prints. The most recent full week:

The cumulative record makes the machinery's track record: roughly $1.1B of Assistance Fund purchases to date — more than 45M HYPE retired — including $379M in 2026 year-to-date through early September. On the burn side, approximately 48.45M HYPE (~4.84% of max supply) has been destroyed between buyback retirement and fee burning. This is the strongest-verifiable part of the whole flywheel: a multi-quarter, multi-billion-dollar audit trail of fees becoming retired supply. Whatever else is uncertain about Hyperliquid, the conversion pipe has receipts.

4. The supply side — two unlocks vs. the bid

A flywheel with a bid but no supply test proves nothing. August and September ran the test twice:

Roughly 24.1M HYPE — about $2.0B of marginal supply — entered circulation in nine days, against a buyback bid running ~$10–13M/week. The absorption was therefore not mechanical: buybacks alone could not have taken down $2B in nine days; price appreciation and new demand did most of the work. The honest conclusion is double-edged. Positive: the flywheel operated in its strongest configuration — rising fees, rising TVL ($5.64B → $6.40B through August), a policy tailwind — and supplyEvents that historically marked local tops (May −14.1%, July −7%) passed without damage. Cautious: the same tape that absorbed the unlocks also faded 12.1% the following week. Absorption under demand is not the same as immunity.

5. The second engine — what AQAv2 adds

The second tap is reserve yield, and its parameters are verifiable even though its output is not yet:

• Activated Aug 26, 2026 — roughly 90% of the yield on the platform's USDC reserves now routes to the Assistance Fund.
• Custody and tech: Coinbase manages the reserve balances; the yield infrastructure runs on Circle's rails.
• First buyback: Oct 3, 2026. Until that print, AQAv2's contribution to the buyback is exactly $0 — verified.
• Modeled size: third-party analysis estimates the stream at ~$527K/day (~$192.5M/year) on ~$6.2B of reservable USDC at ~3.10% rates and 90% routing. We label this Modeled: the rate varies, the reservable base is an estimate, and the first execution will settle the question.

If the model holds, AQAv2 adds a demand stream comparable to 3–4 quiet weeks of trading-fee buybacks per year at roughly $192M — and, critically, one that is counter-cyclical: reserve yield accrues with balances held, not with trades executed. In a week like Sep 7–13 (fees −22% WoW), a steady reserve-yield tap is exactly the kind of ballast the fee engine lacks. The Oct 3 print is the single most important datum for the second-engine thesis this quarter; our weekly data reports will mark it the day it lands.

6. What we cannot measure (yet) — an honest list

The flywheel conversation is full of numbers that do not survive contact with an API. Ours included, historically. The current gaps:

We flag these not as disclaimers but as a research agenda: each row converts a debate into a datapoint once the series exists. Until then, any precise-sounding DeFi-fee or staking figure you see — including in our own archive — deserves skepticism.

7. Scenarios & what to watch

Watch four things: the weekly fee print (our data report carries it), the Oct 3 AQAv2 execution versus the ~$527K/day model, the next unlock's landing (the schedule is public), and whether the transparency proposal converts the ~18% pass-through estimate into disclosed data — every one of these either strengthens or breaks a specific row of this report.

8. Conclusion

Measured honestly, Hyperliquid's flywheel is more mechanical than its critics assume and less complete than its advocates claim. The mechanics are genuinely strong: $67.3M of August fees, ~97% of net capture programmatically converted to buybacks, ~$1.1B of purchases and 48M HYPE destroyed as a standing audit trail, and $2.0B of unlock supply absorbed in nine days without breaking price. The incompleteness is equally real: the fee tap is rate-sensitive and has faded four weeks running; the second tap has produced $0 so far and $527K/day is a model, not a measurement; and the most-quoted number in the ecosystem's DeFi story — the HyperEVM fee base — cannot currently be verified by anyone, including us. The flywheel is real. How many engines power it is the question October answers.

FAQ

Related reading

• Hyperliquid Market Report — August 2026
• Half the Tape? What HIP-3 Builder Markets Really Take and Give
• Hyperliquid Weekly Data Report — Week of August 24–30, 2026
• HYPE Market Intelligence — Week of August 24–30, 2026