Hyperliquid collects 14.57 million USDC via AQAv2: 2.85% net

The first AQAv2 distribution entered the Hyperliquid Assistance Fund at the UTC day close: 14,572,374.16 USDC between 23:40:38 on Oct-3-2026 and 00:00:39 on Oct-4. That amount squares in two ways that arithmetic cannot untie: a net yield of 2.85% annually on the entire on-chain USDC balance, or the Treasury bill rate on a restricted base of $5,164 billion, three-quarters of that balance. The four-week U.S. Treasury bill averaged 3.8145% in coupon equivalent during the accrued thirty days, from Aug-26 to Sep-25-2026, and the average daily USDC balance on-chain was $6,916.7 million. With those two numbers and the declared 90% of net yield, the payout comes to $19.52 million. The measured amount is 74.67% of that: $4.94 million less per cycle. This article clears both readings with the arithmetic in plain sight, recalculates at what level the USDC yield would equal the income that trading leaves to the fund (9.37%, previously 9.82%), measures what the fund bought with that money in the following seven hours, and re-measures Hyperliquid's native lending against HyperLend and Felix twelve days after the last snapshot.

What is the actual yield on Hyperliquid's USDC according to the first AQAv2 cycle?

The published mechanism is a multiplication of four factors: the balance on which it accrues, the rate at which it accrues, the portion that reaches the fund, and the fraction of the year. Of the four, two are published—90% of the net yield and the thirty-day cycle with settlement eight days later—and two are not. A known amount then leaves an equation with two unknowns, and what can be done is to fix one and solve for the other.

The unit convention, declared so that every figure can be recalculated: all rates in this article are simple annual rates, and the window of a cycle enters as 30/365 of a year. A cycle's payout is base × annual rate × 0.90 × 30/365, and when a payout must be compared with an annual flow—the holder revenue, for example—it is the payout that is annualized by multiplying it by 365/30. Mixing the two units in the same line produces errors by a factor of twelve, which is the most common confusion when reading this mechanism. And the naming convention: the distribution and fund balances are given in USDC, which is the unit in which the API returns them and is taken at par with the dollar; chain bases, fee flows, and counterfactual amounts are in dollars.

With the average daily balance of the accrued window—$6,916.7 million, the average of the 31 closes from Aug-26 to Sep-25-2026 published by DefiLlama for USDC on Hyperliquid L1—the net rate that produces the 14,572,374.16 USDC measured in thirty days is 2.8481%. If instead of that balance, the $6,740 million used by CoinGecko in its projection—published on Aug-25-2026, with revenue data closed as of Aug-19—is taken, the net rate rises to 2.9228%. Both readings fit within a narrow band: between 2.85% and 2.92%, from 89 to 97 basis points below the four-week bill for the same period.

A warning about the reference rate, because the Treasury publishes two columns for each term and they do not measure the same thing. The bank discount rate gave 3.88% for the four-week bill on Oct-2-2026 and the coupon equivalent gave 3.95%; the second is comparable to a compound annual yield and is the one we use here and the one we used on Aug-30. Between the two there are 6 to 7 basis points, enough for a correct figure to be mislabeled if they are mixed.

The last row is the alternative reading and should not be discarded: if the accrual occurred at the bill rate, then the base subject to distribution would be $5,164.4 million, 74.67% of the USDC that was on average on the chain. It is the "about 5 billion" figure that has been circulating since May and that our own piece from Jul-3-2026 picked up. The two explanations are arithmetically equivalent and produce the same amount; what is not equivalent is what they imply. A net rate of 2.85% on the entire balance says the custodian retains just under one percentage point before distributing. A base of 5,164 million says that a quarter of the chain's USDC is not in the treasury program.

Having the measured amount to six decimal places does not break the tie between the two readings. What the measurement fixes is the product base × rate, not each factor separately: with the press figure the implicit base came to $5,167.1 million and with the measured one it comes to $5,164.4 million, and that $2.7 million difference does not distinguish between the two explanations. The tie-breaker only arrives when base and rate move in different directions between two cycles, and the first chance to see it is the settlement on Nov-2-2026.

There is a clue to prefer the first. Crypto Briefing describes on Oct-3-2026 "prevailing yields of around 3%" and CoinGecko built its projection applying 3% to the full balance, instead of the bill rate. Whoever calculated the figure circulating today was already using, then, a rate close to 3% on the large base. The measured amount yields $485,746 per day and the CoinGecko base case yielded $498,575: the measured amount is 2.57% below that projection.

How much does the payout deviate from what the Treasury bill would yield on the entire reserve?

The measured payout is 25.33% below what the Treasury bill would have yielded on the entire reserve balance. That is the deviation that matters, and it is best seen against two versions of the same calculation. The first is an internal estimate we wrote on Aug-30-2026 as a working hypothesis and did not publish: the 3.75% coupon equivalent from Aug-28 on the 6,725.3 million USDC on-chain, with the 90% distribution and the 30/365 window, gave $18.66 million. The measured amount is 21.89% lower. The second is the same method with the already closed data for the period, higher in both factors—average balance of 6,916.7 million and average bill of 3.8145%—giving $19.52 million, and there the deviation opens to 25.33%. The error was, therefore, in the rate. Treating the yield of a custodied reserve as if it were the bill coupon overestimates the payout by a quarter.

The condition we had set in writing to consider that estimate wrong was for the payout to fall outside the $14 to $19 million band. The 14,572,374.16 USDC measured falls within, 572,374 USDC from the floor. The band held because it was wide precisely due to not knowing the base. The specific figure of $18.66 million failed by $4.08 million.

The useful residue of that failure is the net rate band. If the distribution is 90% of the net yield and the net is around 2.85-2.92% when the four-week bill is at 3.81%, then the custody channel keeps something close to one percentage point. Neither the gross rate nor the distribution of that difference are published: the framework documentation sets 90% "of the net yield" and leaves undefined what is deducted to reach the net, so the only available value is the 2.85-2.92% solved from the amount.

When did the first AQAv2 distribution enter the Hyperliquid Assistance Fund?

It entered at the UTC day cut-off. The Hyperliquid Assistance Fund had 10,852.65 USDC at 19:58 UTC on Oct-3-2026, 11,102.57 six minutes later, 21,344.90 at 20:20, a peak of 27,009.30 at 20:40, and 9,765.79 at 23:40:38. Twenty minutes later, at 00:00:39 UTC on Oct-4, it had 14,582,139.95. The time matters at both ends of the jump. During the three hours and forty-two minutes of readings on the day the press considered settled, the fund had nothing in its spot account above those 27,009.30 USDC: anyone looking at the balance that afternoon would have seen an empty account and a press figure with no on-chain backing. And the money appeared just as it crossed midnight UTC, when that settlement day ended.

The effective amount is 14,572,374.16 USDC, and the method by which it emerges is the jump between two consecutive readings of the fund's spot balance: 9,765.79 at 23:40:38 UTC on Oct-3 and 14,582,139.95 at 00:00:39 UTC on Oct-4. The next reading, at 00:20:41, gives 14,591,777.63 and confirms the level. Nor is it an entry and exit within the sampling window: the series continued taking the balance every twenty minutes for seven hours and twenty minutes more, and the account stayed in the 13.19 to 14.60 million USDC band, with the last reading, at 07:21:04 UTC, at 13,187,635.29. Between the first reading with the money inside and the last of the series, there are 7 hours and 20 minutes and twenty-one intermediate readings, all above 13.3 million USDC.

The fund's movement ledger would not have served to see it, and that reinforces the decision to measure it by balance. Its 646 entries since Oct-15-2025 are of only two types, sends and spot transfers; the entry of trading fees does not appear there and the AQAv2 distribution left no trace either. An entry that reaches the spot balance without passing through that ledger is only seen in two ways: by looking at the balance with sufficient frequency or by looking at what the fund does with it next.

The latter was also measured, and the buying pace changed. The fund bought HYPE for $2.179 million on Sep-30, $1.755 million on Oct-1, $2.502 million on Oct-2, and $0.546 million on Oct-3 until 20:03 UTC; these are the $1.8 to $2.5 million daily that alone explain the September fees, because $55.89 million distributed over thirty days gives $1.863 million a day. Between 20:20 and 23:40 UTC on Oct-3, before the deposit, the fund's cumulative acquisition cost rose $84,643.29 in three hours and twenty minutes: an equivalent of $0.61 million a day. Between 00:00:39 and 07:21:04 on Oct-4, with the money inside, it rose $1,524,537.28 in seven hours and twenty minutes: an equivalent of $4.98 million a day, with 16,991.30 HYPE bought at an average of $89.72. The pace the morning after the distribution doubles that of the most active day of the previous week.

The money is already being spent. At 07:21:04 UTC on Oct-4, 13,187,635.29 USDC remained in the account, equivalent to 90.50% of the distribution; the balance dropped $1,394,504.66 since the first reading with the money inside while the acquisition cost rose $1,524,537.28—the $130,032 difference is the trading fee that continues to enter in parallel. At the buying pace of those seven hours, the distribution is exhausted in 2.65 days; at the average September pace, in 7.1.

The press figure is confirmed by an independent route, and it is best not to inflate what that means. Crypto Briefing published $14.580 million on Oct-3 and the amount measured in the account is 14,572,374.16: a difference of $7,626, or 0.05%. Whoever passed the number to the press was already working with the exact amount, a day before the fund balance showed it. What still does not exist is an announcement from the protocol, so the primary source for this figure is the fund balance, not the statement.

There is a second cross-check, over a long stretch, which is what gives confidence in the aggregator's series. The fund's cumulative acquisition cost—the literal label of the entryNtl field from the Hyperliquid API, which accumulates what the fund has paid for the HYPE it holds, valued at the price of each purchase—went from $1,267,674,506.81 on Aug-30-2026 to $1,334,390,918.58 on Oct-3. That is $66.72 million spent in 34 days. The holder revenue that DefiLlama attributes to Hyperliquid in that same stretch totals about $63.39 million. The deviation is 5.25%, and it is the verification that the aggregator's series reproduces what the fund does with its money.

How does AQAv2 weigh against what trading leaves in the Assistance Fund?

September 2026 left $55.89 million in holder revenue—the label the aggregator uses for the portion of fees that goes to the buyback fund—out of $72.18 million in total protocol fees. Adding the measured AQAv2 distribution, the fund's monthly income rises to $70.46 million, of which the reserve yield contributes 20.68%. It is one-fifth, and it is the first fifth that does not depend on anyone trading. Taken to a year—the cycle is thirty days, so it is multiplied by 365/30—the distribution is equivalent to $177.30 million annually, 30.40% of the $583.16 million that the annualized Q3 2026 holder revenue comes to. Annualization is a scale for comparison: it assumes that the balance and the rate stay where they are.

Two figures in that table need precision regarding what they measure. The first is ours: the internal estimate from Aug-30 noted $47.97 million for August, which was the unclosed month read that same day; closed it gives $51.48 million, 7.3% more. The second we have repeated in at least five of our own articles, starting with the one on Hyperliquid's revenue fundamentals: the "97% of fees" that goes to the fund. Measured as a ratio of the two aggregated series, what reached the fund was 76.46% in August and 77.43% in September. The 97-99% is correct over another denominator—perpetual fees minus builder fees, according to the methodology declared by the DefiLlama adapter—narrower than the aggregated fee series. They are two ratios with two different denominators, and each is correct over its own.

The context that matters for reading the 20.68%: fees are shrinking. September 2026 holder revenue is 34.40% below that of September 2025, and the full quarter is 49.70% below Q3 2025. A new income stream that does not depend on volume weighs more every month that volume drops, and weighs less every time the Fed cuts.

At what rate would the USDC yield equal Hyperliquid's holder revenue?

Our Aug-30 estimate placed parity—the reserve yield equaling the income trading leaves to the fund— at a rate of 9.82%. That threshold drops today to 9.37%, and the cause of the drop is not the Fed's hike.

The threshold is a ratio: annual holder revenue divided by base times distribution. It does not contain the prevailing rate. With the Q3-2026 holder revenue taken to a year—145.79 × 4 = $583.16 million, which is the portion of fees reaching the fund; total quarterly fees, annualized, give $778.56 million—and the average base of the accrued window ($6,916.7 million) at 90%, it comes to 9.3680%. It dropped 45 basis points compared to 9.82% because the base grew 2.85% and that income fell 1.92% relative to the previous quarter, from 594.6 to 583.16 million. What the Fed hike of Sep-16-2026 moved, which took the target range to 3.75-4.00% in a 12-0 vote, is the distance to the threshold: it was 607 basis points on Aug-30 (9.82% vs. 3.75% bill) and is 542 on Oct-2 (9.37% vs. 3.95%). The gap narrowed by 65 basis points.

The last row is the one that rules if the 2.85% net rate is the correct explanation. What the fund actually receives is 74.67% of what 90% of the base would yield at the bill rate, so to reach parity the payout would have to multiply by 3.29: a four-week bill of 12.55%, or an on-chain USDC balance of $22.75 billion, or a 69.60% drop in holder revenue from the Q3 2026 level. All three are counterfactuals of the same ratio, posed to dimension it.

Sensitivity is also recalibrated downward. With the full 90% on the balance, every 100 basis points of rate moved $60.5 million a year. With the measured pass-through, they move $46.48 million, 7.97% of the annual holder revenue. For the reserve yield to equal that revenue, another 860 basis points of bill rate above the 3.95% of Oct-2-2026 would be needed.

What does the Assistance Fund do with the HYPE it buys?

The fund does not burn the HYPE it buys. It keeps it in its own balance. It had 47,657,692.75 HYPE on Oct-3-2026, compared to 46,885,580.63 on Aug-30, and 47,675,794.47 at 07:21:04 UTC on Oct-4, already spending the distribution. The 772,112.12 new units from the stretch remain in its balance. The average cost of the entire stock rose from $27.04 to $28.00 per HYPE, because the new stretch was bought at an average of $86.41 per HYPE. At the $89.299 mark on Oct-3, that stock is worth $4,255.8 million.

What decides the reading of those units is how the protocol counts them. The Hyperliquid API expressly lists the fund address among non-circulating balances, alongside the future emissions address, the zero address, and the burn address. That is: the circulating supply of 298,637,420.40 HYPE on Oct-3 excludes the fund's stock. Saying the fund holds "15.96% of the circulating supply"—our Aug-30 estimate put it at 15.70%—is dividing two sets that do not overlap, and the resulting ratio does not measure any share. What is a share: the fund holds 4.77% of the maximum supply of 1,000 million HYPE.

And regarding burning, the only number the API publishes is the gap between the maximum supply and the declared total supply: 1,111,934.59 HYPE on Oct-3-2026, 0.11% of the maximum. The API does not label that gap, so no mechanism can be attributed to it. The 772,112.12 units bought since Aug-30 fall outside that gap and remain accounted for in the fund balance.

What does the AQAv2 calendar mark from the June vote to the November cycle?

The cycle is the mechanical piece most misunderstood, because the money from the first distribution is from August and September and arrives in October. The accrual runs thirty days and settlement falls eight days after the period close.

The second cycle accrues on a larger balance—the chain had $7,309.8 million in USDC on Oct-3, 5.68% above the first period average—and with the bill at 3.95%. If the measured 74.67% pass-through holds and the bill stays where it is, the Nov-2 distribution comes out to around $15.95 million. That is the number that will verify if 2.85% was the rate or if it was the base: above it, the net rate rules; clearly below it, the restricted base rules.

What happened to Hyperliquid's native lending against HyperLend and Felix?

The Sep-22-2026 piece left three written conditions to be proven wrong and set Oct-3 as the verification day. None were met, and the reason none were met is the same price the piece described.

• Native USDC utilization above 85% for three consecutive days with funding at the floor: No. The reserve was at 74.82% at 20:03 UTC on Oct-3 and 74.91% one minute later, 10.18 percentage points below 85%.
• Lent USDC growing more than 10% on days with average funding below 5% annually: Not evaluable day-by-day, as the protocol API does not store utilization series. What is measurable is the full stretch: lent amount rose 11.21% in the twelve days from Sep-21 to Oct-3, from $354.76 to $394.54 million, with BTC, ETH, and HYPE funding at the exact floor in both daily readings.
• HyperLend stablecoin book more than 25% below the $67.33 million from Sep-21: No. It was at $53.86 million, 20.01% below. The margin to the threshold was $3.36 million.

The two prices set by the protocol remain exactly where they were: native lending at 5,000% across all five reserves and the funding floor at 0.0000125 per hour, 10.95% annualized, in BTC, ETH, and HYPE. The structural spread of 595 basis points between the two has not moved in twelve days. There was an extra check that was not planned: SOL fell below the funding floor, to 1.67% annualized, with a premium of −0.000542 against the oracle. The negative premium exceeded the −0.04% cap that the formula absorbs, which is the only way the floor stops applying, and is direct verification of the mechanism we described on Sep-22.

HyperLend's drop is not a one-off reading within a back-and-forth movement, and this can be stated because there are thirteen daily closes in between. The lent series published by DefiLlama drops monotonically from $275.12 million on Sep-21 to $209.17 on Oct-3 without a single upward reversal, and matches the primary reading from the protocol's own API ($209.03 million) with a 0.07% deviation. In the same period, HyperLend cut its USDC reserve caps for the second time: the supply cap dropped from 77.36 to 64.60 million and the lending cap from 70.08 to 55.57 million, and the rate curve remained unchanged since Sep-21. Its USDC lending rate was at 4.788% with 87.76% utilization, below the native 5,000%.

The WHYPE rate on HyperLend did move sharply, from 17.840% to 0.994%, and the cause was not a parameter change: utilization crossed below the 80% optimal, to 79.51%, and the curve returns to a flat slope there. The price itself did the work the Sep-22 piece anticipated on the depositor side.

What changes for HYPE holders since the first AQAv2 distribution?

What changes is the composition of income, not its size. The fund that repurchases HYPE now collects from two places: $55.89 million from trading in September and $14.57 million measured from reserve yield, with the latter contributing 20.68% of the monthly total. That one-fifth is insensitive to volume and sensitive to two things the protocol does not decide: what the Fed does with rates and what the custody channel retains before distributing. The framework documentation sets the distribution at 90% of the net yield without publishing the gross rate, so those 89 to 97 basis points only emerge by solving for them from the amount.

Three things remain measurable with dates for anyone wanting to follow the second cycle with their own readings. The first is the spending of the distribution already inside: at 07:21:04 UTC on Oct-4, 13,187,635.29 USDC remained of the 14,572,374.16 received, and the buying pace of those seven hours equals $4.98 million a day compared to $1.8 to $2.5 in previous weeks; if the pace returns to its average, the first cycle money is exhausted before the second arrives. The second is the Nov-2-2026 cycle, which accrues on a 5.68% higher balance and with the bill at 3.95% on Oct-2, 13.5 basis points above the 3.8145% average of the first period: if the distribution does not rise in that proportion, the net rate explanation falls and the restricted base one gains. The third is the distance to the parity threshold, 542 basis points on Oct-2, which narrows via two routes—rates up or fees down—and which at this rate of volume contraction will move more by the second than the first.

Sources and links: Hyperliquid API, info endpoint (spotClearinghouseState, tokenDetails, allBorrowLendReserveStates, metaAndAssetCtxs, userFillsByTime, userNonFundingLedgerUpdates) · Hyperliquid Documentation, portfolio margin (rate curve and caps) · U.S. Department of the Treasury, daily bill rates (bank discount and coupon equivalent columns) · Federal Reserve Bank of New York, reference rates · DefiLlama, USDC series on Hyperliquid L1 · DefiLlama, Hyperliquid holder revenue series · DefiLlama, declared methodology for Hyperliquid perpetuals adapter · HyperLend API, markets on HyperEVM · DefiLlama, Felix · Crypto Briefing, first AQAv2 distribution amount and validator approval · CoinGecko, AQAv2 projection published Aug-25-2026 with data to Aug-19 · PANews, attribution of the figure to the HL HUB community account · Federal Reserve, FOMC calendar and decisions