HYPE tokenomics: the Assistance Fund buyback is a claim on volume, not a floor

The standard $HYPE story is that it is the buyback token: roughly 99% of Hyperliquid's fees are converted into HYPE and taken out of supply, so the float only shrinks. The story is half-right, and the half that gets repeated is the half that is easy to verify.

The half that gets skipped is what the buyback is actually a claim on. A protocol that spends 99% of its fees buying its own token has not created a price floor. It has created a mechanical, non-discretionary bid whose size is a linear function of trading volume, sitting opposite a supply schedule where 38.888% of the token has never been issued. HYPE is a levered claim on Hyperliquid's fee throughput racing an emission calendar. This piece pins the exact fee split from the primary source, separates the December 2025 burn vote from the fee rule it did not touch, and shows where that fee engine is visible on QuantAbundancia's own market board.

What the Assistance Fund actually is

The distinction that matters is between a buyback and an execution rule. Most token buybacks are treasury operations: a foundation holds fees, decides when to buy, and can stop. The Assistance Fund is neither discretionary nor off-chain. Trading fees are converted into HYPE as part of Hyperliquid's L1 execution and routed to a designated system address, 0xfefefefefefefefefefefefefefefefefefefefe, which was created without any control mechanism for withdrawal.

The practical consequence is that nobody schedules these purchases and nobody can pause them. The bid appears because trades cleared. It is closer to a protocol tax denominated in its own token than to a corporate repurchase authorisation, and that structural difference is why the mechanism keeps running through drawdowns, which is exactly when a discretionary treasury stops buying.

Two smaller inflows attach to the same address. USDH, the Hyperliquid-native stablecoin issued by Native Markets, routes 50% of its reserve yield to the fund, where it converts on the same path. That contribution is not fee-driven, so it moves with rates and float rather than with volume.

The exact split, and the one date it changed

This is where most secondary coverage rounds to "99% of fees" and loses the two things a reader actually needs: the rule is not the same on both venues, and it changed once. The primary source is Hyperliquid's own fee documentation, which states it directly:

Hyperliquid Perps: 99% of fees go to Assistance Fund for buying HYPE tokens, excluding builders fees.

Hyperliquid Spot Orderbook: 97% of spot trading fees before 30 Aug 2025 and 99% thereafter go to Assistance Fund for buying HYPE tokens, excluding unit protocol fees.

Read against the question "does the rule differ between perps and spot", the answer is yes on two counts, only one of which has closed:

The headline percentages converged in August 2025. The carve-outs did not, and they are the part that matters going forward, because they are the leak in the model. Builder fees are capped at 0.1% of notional on perps and 1% on spot, and they are deducted before the fund sees anything. Any growth in third-party front-ends routing order flow is growth in a line item that sits ahead of the buyback in the waterfall, not behind it.

The burn is a claim on volume, not a floor

A fixed percentage of a variable base is still a variable. The buyback's dollar size is set by fee revenue, and fee revenue is set by volume and by the fee schedule, currently 0.045% taker and 0.015% maker on perps, 0.07% taker and 0.04% maker on spot, before volume tiers.

Reported figures put cumulative Assistance Fund purchases above $1.3B since launch, averaging around $1M per day, an annualised rate somewhere near 7% of market capitalisation. Those numbers come from third-party trackers rather than a protocol disclosure, so the order of magnitude is the usable part, not the decimal.

The framing that survives is this: the mechanism converts trading activity into supply reduction at a near-fixed rate, which means the token's deflation is procyclical. Volume expands, the burn expands. Volume contracts, the burn contracts, and it contracts precisely when the emission schedule below is still running on its own clock. A floor implies a bid that does not depend on conditions. This one depends entirely on conditions.

The other side of the ledger: 38.9% of supply has never been issued

Total supply is fixed at 1,000,000,000 HYPE. There was no private sale and no venture allocation at launch, which genuinely inverts the usual L1 cap table and is the strongest thing in the distribution's favour. The genesis breakdown:

The airdrop was liquid on day one, which is why the float was large immediately. The two lines that matter for the next two years are the ones that have not moved yet.

Future emissions, 38.888%. This is the largest single allocation and it has no published release schedule. It has historically funded validator incentives and community rewards. An unscheduled reserve of that size is not a reason to discount the burn, but it is the reason the burn cannot be read as a one-way supply function.

Core contributors, 23.8%. Cliffed into 2027 and 2028, then released monthly on the 6th. The detail worth carrying: the actual distribution is discretionary. The team announces a claim amount each month, and the announced amount has consistently run well below the theoretical maximum. This is why circulating-supply figures disagree so widely, with vesting trackers showing roughly 222M HYPE unlocked and other aggregators counting closer to 333M. The gap is methodology, not error, and anyone quoting a market cap for HYPE is quoting one of those two conventions without saying which.

What the December 2025 vote did, and what it did not do

These two things get merged constantly, including in coverage that should know better, so it is worth separating them cleanly.

The validator vote announced on 17 December 2025 asked validators to formally recognise the HYPE already sitting in the Assistance Fund address, roughly 37M tokens or over 13% of the circulating supply at the time, as burned and therefore excluded from circulating and total supply. Validators signalled on the governance forum by 21 December at 04:00 UTC, stake could move to a matching validator until 24 December at 04:00 UTC, and the result was taken as stake-weighted consensus at that timestamp. It reportedly passed with 85% of stake in favour, 7% against and 8% abstaining.

What it changed is accounting and governance treatment, not economics. The tokens were already unreachable without a hard fork; no on-chain action was required, and the vote did not reduce supply that was otherwise going to circulate. What it did not change, at all, is the fee percentage. There is no validator vote that "set" the 99%. The 97% to 99% spot move on 30 Aug 2025 is a protocol parameter recorded in the documentation, not the output of a stake-weighted governance decision. If you are looking for a proposal that fixed the buyback share, it does not exist in that form.

Where the fee engine shows up on QA's own board

The abstraction becomes concrete at the point where Hyperliquid's fee base overlaps the equity map QuantAbundancia already maintains. HIP-3 lets third parties deploy their own perp markets by staking 500,000 HYPE, and the deployers have used it to list equities. QA's collector reads that surface every five minutes.

As of 2026-08-29 21:47 UTC, 47 QA instruments have a live Hyperliquid perp, carrying $159.4M of 24-hour volume against $1.75B of open interest. 44 of the 47 sit on the xyz sub-dex, which is the only third-party venue with meaningful equity listings.

The overlap is not incidental. 36 of those 47 markets map to a QA bubble, and those 36 carry $102.7M, or 64%, of the volume on the board. The concentration is in exactly the complex the site was built around: $HXSCL (SK Hynix) is the largest at $36.9M, followed by $SNDK at $13.6M and $MU at $10.0M, all three in the DRAM / HBM memory bubble, with $NVDA at $7.6M in semiconductors.

The relevance to tokenomics is direct. Every one of those fills pays a fee, 99% of that fee net of the builder carve-out converts to HYPE, and the resulting bid is funded partly by leverage taken on the same names QA maps on the equity side. The board is at /hyperliquid. If you want to see the venue mechanics before reading about them, /stack/hyperliquid covers how the exchange works, including a bot of ours that failed on it and why we published the post-mortem.

What to watch

• Builder-fee share of notional. The carve-out sits ahead of the buyback in the waterfall. Third-party front-ends already account for a large share of daily active traders, and growth there is a direct deduction from the fund's inflow.
• Monthly core-contributor claims, on the 6th. The announced amount versus the theoretical maximum is the cleanest read on whether the discretionary restraint holds as the cliffs open through 2027 and 2028.
• Any published schedule for the 38.888% emissions reserve. It has none today. A schedule would convert the largest unknown in the supply model into an arithmetic one.
• Fee revenue against unlocks, monthly. This is the whole question in one ratio: dollars of fees converted to HYPE versus HYPE newly released. The burn narrative holds while the first outruns the second and stops being a supply story when it does not.
• HIP-3 equity volume as a share of total perp volume. If the equity sub-dex keeps compounding, the fee base behind HYPE becomes increasingly a function of the AI and memory complex rather than of crypto beta, which is a different exposure than most HYPE holders think they have.

Bubble shifts and rule-based alerts on the names carrying this volume are part of /pro.

Live venue data: /hyperliquid - every QA instrument with a live perp, its mark, funding and open interest, refreshed every five minutes.

Bubble context: /bubbles/memory - the cluster carrying the largest share of equity perp volume on the board.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.