Hyperliquid lends at 5% and its funding pays shorts 10.95%

Hyperliquid lends USDC at 5.00% annually while simultaneously paying shorts on any of its perpetuals a funding floor of 10.95% annually, provided the contract trades flat against the oracle. Both numbers are set by the protocol. The first is the rate curve for native portfolio margin lending (the account mode of HyperCore, Hyperliquid's trading layer, which margins spot and perpetuals together and automatically borrows USDC against collateral). The second is the interest component of funding (the periodic payment between longs and shorts that anchors a perpetual—a future without expiry—to the oracle, the external reference price the protocol uses as spot). In the 735 hours between Aug-22 and Sep-21, 2026, BTC funding was at that floor 70.5% of the time, ETH 79.2%, and HYPE 69.7%. A covered short (cash-and-carry: selling the perpetual while holding the same asset in spot; price ceases to matter and only funding remains) that financed the entire notional of its position using the native book yielded between 4.84 and 5.24 net points in that window. The curve crosses the funding floor at 81.25% utilization; the USDC reserve was at 72.55%, $42.52 million away from that ceiling.

What two prices does Hyperliquid set in its native portfolio margin lending and funding?

The portfolio margin documentation defines the lending rate for stablecoins with a single formula: 0.05 + 4.75 × max(0, utilization − 0.8), in continuous compounding and indexed hourly to match the funding interval. Up to 80% utilization (borrowed USDC divided by supplied USDC), the rate is flat at 5.00%; beyond that kink, it rises by 4.75 points for every point of utilization. On Sep-21, 2026, at 22:05 UTC, the USDC reserve returned exactly that: a 5.000% lending rate with 72.55% utilization. The USDT0 reserve (the version of USDT circulating on Hyperliquid) provided a slope check on Sep-19, 2026: at 12:11 UTC, it was at 80.17% utilization and charging 5.798%; seventeen basis points of utilization above the kink are worth 0.8 rate points. By 14:31 that day, the USDT0 supply had grown by 4,725 units in 2h 20min, bringing the reserve back below the kink (79.98%, 5.000% rate); the borrowed amount did not move. The curve dropped due to the supply side: 4,725 USDT0 entered, and the borrowed amount remained at 1.59 million. Two days later, at 22:05 UTC on Sep-21, that reserve was at 41.66%: supply had risen to 3.01 million and borrowed funds had dropped to 1.25 million.

The funding documentation sets the second price: F = P + clamp(0.01% − P; −0.05%; 0.05%), where P is the premium (the average difference between the perpetual's impact price and the oracle, averaged hourly) and 0.01% every 8 hours is the interest component, "defaulted for consistency with centralized exchanges," paid in eighths every hour; the full calculation is in the explanation of funding in perp DEXs. With the perpetual flat against the oracle, P = 0 and the long pays the short 0.00125% per hour on the notional at oracle price, even if there is no imbalance between buyers and sellers. Annualized simply, this is 10.95%; the documentation lists "11.6% APR," which matches the rate compounded hour-by-hour (1.0000125 to the power of 8,760 equals 11.57%). Here, the simple 10.95% is used because funding is collected hour-by-hour without automatic reinvestment and because the lending curve, although it is APY in continuous compounding (5.00% equals 4.88% nominal), sits within 0.12 points of the same base.

The difference, 5.95 points at zero premium, stems from two constants in the documentation and remains as long as utilization stays below 80%.

How much did the carry yield on Hyperliquid between Aug-22 and Sep-21, 2026?

The floor only matters if funding spends its life there. The fundingHistory hourly series, 735 records per coin from Aug-22, 2026, at 08:00 UTC to Sep-21 at 22:00, confirms this for the platform's three largest open positions. BTC was exactly at 0.0000125 per hour for 518 of those hours (70.5%), ETH for 582 (79.2%), and HYPE for 512 (69.7%). The longest streak without leaving the floor was 268 hours for ETH, eleven straight days between Aug-28 and Sep-8, 2026. Negative hours, where the short pays, occurred 47 times for BTC, 37 for ETH, and 79 for HYPE.

The net carry in the table is a proprietary and conservative calculation: funding collected on a notional minus the 5% interest on that same notional during the same 735 hours (interest totals 0.4195% for the period). In BTC, the accumulated funding was 0.8280%, leaving 0.409%, which annualizes to 4.87 points. HYPE is the coin with the highest premium and also the one that most frequently falls below the profitability threshold: 132 hours, or 18.0% of the period, where hourly funding paid less than the prorated 5% annual rate (0.00000571 per hour). Its hourly peak, 0.000152 on Sep-17, 2026, at 14:00 UTC, is equivalent to 133% annualized during that hour.

The window was not uniform. From Aug-22 to Sep-6 (376 hours), annualized funding was 11.07% for HYPE, 11.31% for BTC, and 10.45% for ETH, with ETH at the floor 90.2% of the time. From Sep-7 to Sep-17 (264 hours), it dropped to 6.77%, 6.83%, and 7.97%: perpetuals traded below the oracle more frequently, and the clamp stopped covering the negative premium beyond −0.04% per interval. This was the only window where the carry approached the cost of borrowing, and it still outperformed it by 1.8 to 3.0 points. From Sep-18 at 00:00 until the Sep-21 snapshot (95 hours), it returned to 16.62%, 12.61%, and 12.58%.

How is the covered short set up with Hyperliquid native lending?

The portfolio margin documentation provides the example by name: "the carry trade." A user holds 1 BTC in spot and opens a 1 BTC short in the BTC-USDC perpetual with 10× leverage. The margin for the short is automatically provided by native lending, using the BTC as collateral (LTV 0.50: every dollar of BTC allows borrowing up to $0.50 of USDC). Gains and losses from spot and perpetuals are offset in the same account, so a rise in BTC does not liquidate the short: the protocol borrows more USDC against the BTC, which is now worth more. Generic cash-and-carry is explained in the guide to delta neutral strategies.

The arithmetic using the oracle from Sep-21, 2026, at 22:05 UTC ($86,692 per BTC) is as follows. The 1 BTC short collects, at the floor, 10.95% on $86,692: $9,493 per year, in hourly payments of $1.08. With 10× leverage, the borrowed margin is $8,669, and 5.00% on that figure costs $433 per year. If the user borrows the entire notional, the interest rises to $4,335 and the net remains at $5,158, or 5.95% of the notional. The 4.84-5.24 points in the 735-hour table are calculated under this second, more conservative assumption; with margin at 10×, the net on equity is higher, as is the liquidation risk. In the 30.6 days measured, that 1 BTC short would have collected $718 in funding and paid $364 in interest on the full notional ($36 if only financing the 10× margin).

At what utilization does Hyperliquid native lending cease to be carry?

The 4.75 slope causes the margin to evaporate in just over one point of utilization. Equating the curve to the funding floor, 0.05 + 4.75 × (u − 0.8) = 0.1095, yields u = 81.25%. Above this value, borrowed USDC costs more than the funding pays at zero premium, and the covered short financed in the native book loses money by design. With the $488.95 million supplied on Sep-21, 2026, 81.25% equates to $397.27 million borrowed: $42.52 million more than the $354.76 million in the snapshot. This is the demand ceiling for the book if the only demand were carry at floor funding; between the reading on Sep-19 at 14:31 UTC and Sep-21 at 22:05, borrowed USDC grew by $50.73 million in 55.5 hours.

The curve punishes high utilization much more severely than HyperLend, which on Sep-21, 2026, reached 4.80% at its 90% optimum and only spiked thereafter (14.80% at 95%). Price itself pushes native utilization back below 80%: a borrower seeing the 28.75% at 85% utilization either returns USDC or closes the short, and a depositor seeing it supplies more. If the USDC reserve stays above 85% with funding at the floor, it would signal that demand is coming from elsewhere, which is what the falsification test monitors.

What can liquidate a carry on Hyperliquid with portfolio margin?

The carry breaks in four ways, none of which require the market to move against BTC:

1. Negative funding: 47 of the 735 hours in BTC and 79 in HYPE between Aug-22 and Sep-21, 2026.
2. USDC reserve utilization above 81.25%, where borrowing costs more than the funding floor.
3. Portfolio margin liquidation upon exceeding a 0.95 ratio, with collateral thresholds of 0.75 for BTC and 0.825 for HYPE.
4. Collateral caps of 2,000 BTC and 10M HYPE: HYPE was at 118.2% on Sep-21, 2026; BTC at 99.7%.

The first is funding itself: 47 of the 735 BTC hours were negative, and 79 for HYPE. Since the clamp limits the interest component to offsetting a negative premium of up to 0.04% per interval, the 8-hour funding falls below 5% annual (0.00457% per interval) only if the perpetual trades on average more than 0.0454% below the oracle; in HYPE, this occurred 18.0% of the hours in the period.

The second is the rate curve from the documentation: if other borrowers push utilization above 81.25%, the lending rate exceeds the funding floor even if the market is flat. The third is liquidation. Portfolio margin margins cross perpetuals (those sharing margin instead of having it isolated per position) together with the account's spot balances; the account becomes liquidatable when its maintenance ratio exceeds 0.95. The collateral liquidation threshold is 0.5 + 0.5 × LTV, 0.825 for HYPE and 0.75 for BTC. In the documentation example, with BTC at $200,000, the user "must reduce the carry notional to avoid a loan liquidation." There is no market phase: the system's backstop liquidator (address 0xbbb…b) takes collateral and debt in 20% tranches and converts them to USDC with a 10-minute half-life TWAP, and the order of oracle updates decides whether the perpetual or the loan is liquidated first, "and users should not expect a deterministic sequence."

The fourth are the caps. The documentation sets 10M HYPE and 2,000 BTC as global collateral, and the API on Sep-21, 2026, returned 11,815,721 HYPE and 1,994 UBTC (Unit's wrapped bitcoin, Hyperliquid's asset gateway) supplied: 118.2% and 99.7% of those caps. BTC supply has returned below its cap since the 2,073 UBTC on Sep-19; HYPE exceeds it by 1.8 million units. Either the documentation lags behind the live parameter or the cap does not count what enters via another route; in both cases, "portfolio margin accounts revert to non-portfolio margin behavior when caps are reached," and once the lending cap is hit, additional margin must be provided by the user in the settlement asset (USDC). Collateral is worth $1,275.0 million at oracle price ($1,102.1 million in HYPE and $172.9 million in BTC) and allows borrowing up to $802.8 million by LTV; the $356.03 million borrowed across USDC, USDT0, and USDH uses 44.3% of that capacity, and the $354.76 million in USDC uses 71.0% of its $500 million global cap.

Where does the 10% interest retained by Hyperliquid go?

The documentation is literal: "the protocol retains 10% of the interest from loans as a buffer for future liquidations." It is the mechanic of the reserve factor in Aave or Compound but with a different destination: a liquidation reserve, not a treasury. The Assistance Fund, which repurchases HYPE, is funded by trading fees and not this interest (the full circuit is in the analysis of Hyperliquid revenue). The API confirms the retention with the deposit rate: 5.00% lending × 72.55% utilization × 0.9 = 3.2649%, the exact supplyYearlyRate figure. The native depositor earns 3.26%.

The 5.95 carry points are paid by the perpetual long, not the protocol. On the $3,813.3 million of BTC open interest, $3,191.1 million of ETH, and $2,026.8 million of HYPE (API units at mark price, the perpetual's internal reference price, at 22:05 UTC), every hour at the floor transfers 0.00125% from longs to shorts: in BTC, that is $47,700 per hour and $418 million per year if funding never moved from the floor.

How does Hyperliquid's native book compare to HyperLend and Felix?

HyperLend and Felix live on HyperEVM, Hyperliquid's smart contract layer (the division between HyperCore and HyperEVM is in the architecture analysis). The portfolio margin documentation states that native lending "intentionally does not bring a full credit market to HyperCore" and that it introduces "organic lending demand" for EVM protocols. The correct comparison is between borrowed stablecoins, because the native book only lends USDC, USDT0, and USDH (Hyperliquid's native stablecoin), and because 71.0% of HyperLend's debt ($190.45 million of $268.20 million) is denominated in WHYPE at 17.84%, a HYPE-against-HYPE loan that does not compete with the above. Felix combines lending vaults with a CDP (collateralized debt position: the user locks collateral and mints a stablecoin against it).

The native book lends 5.29 times more stablecoins than HyperLend: $356.03 million versus $67.33 million on Sep-21, 2026. In the 55.5 hours separating that reading from the one on Sep-19, the three books moved in opposite directions: native borrowed USDC rose by $50.73 million, HyperLend's total borrowed on DefiLlama dropped from $315.33 million (Sep-19 at 00:00) to $269.68 million (Sep-21 at 21:14) and its stablecoin book from 95.81 to 67.33, and Felix's TVL went from 93.11 to 83.83 million. Fifty-five hours do not make a series, and in those same hours, HyperLend's USDC curve changed parameters—base from 0 to 4.3%, slope 1 from 7.51% to 0.50%, slope 2 from 55.2% to 20%—altering the price at which it competes regardless of where the money goes. DefiLlama is an aggregator, and these borrowed fields are indeed balances, unlike its open interest, which sums flows, as explained in the guide to reading DefiLlama open interest. Felix publishes TVL by product, but a primary API with its USDC lending rate has not been located, which is why its row remains without rates.

The native book finances trading margin with HYPE and BTC at LTV 0.65 and 0.50, without a debt token and without composability; HyperLend lends USDC against WHYPE at LTV 0.65 with a liquidation threshold at 81.47%. Who retains which tranche of margin in major on-chain credit markets is measured in the on-chain credit retention ranking.

How is the carry thesis on Hyperliquid falsified before Oct-3, 2026?

Status as of Sep-22, 2026 (last reading, Sep-21 at 22:05 UTC): USDC reserve at 72.55% utilization and 5.00% lending; hourly funding for BTC, ETH, and HYPE at the floor (0.00125%/h); HyperLend with $67.33 million in borrowed stablecoins and its USDC curve reparameterized. Daily readings until Oct-3, 2026.

The central limitation is that allBorrowLendReserveStates returns the current state and does not store history. This article demonstrates where the two parameters are and how funding behaved over 735 hours; whether utilization rose or fell with funding remains unproven, as no public utilization series exists to cross-reference it. Therefore, 81.25% is presented as a predictive threshold rather than an observed correlation. The automatic book predates the snapshot by nine months: portfolio margin launched in pre-alpha on testnet on Dec-12, 2025, and on mainnet on Dec-23, 2025 (with a $1 million global USDC lending cap), moved to alpha on Mar-10, 2026, expanded limits in beta on Jun-25, 2026, and opened manual borrowing on Sep-18, 2026—the day Hyperliquid posted on X "$269 million in assets borrowed today," the only point prior to the snapshot and without a supporting series.

The test is scheduled with daily readings from both APIs from Sep-22 to Oct-3, 2026. The thesis is false if USDC reserve utilization stays above 85% for three consecutive days with the daily average funding of BTC and HYPE at or below the floor (lending would cost 28.75% or more against 10.95% funding, yet demand would persist: it would not be carry). It is also false if borrowed USDC grows by more than 10% on days when the average funding of HYPE and BTC is below 5% annual, which occurred in 18.0% and 11.2% of the hours in the measured period. And it is false in its comparative part if by Oct-3 the borrowed stablecoin book of HyperLend has fallen by more than 25% from the $67.33 million on Sep-21: between the 19th and 21st, it already fell 29.7% from $95.81 million, during the same hours its USDC curve changed parameters, so the test's baseline is the reading on the day of publication and not the 19th. If the native book stalls near $397 million of borrowed USDC while funding remains at the floor, the prediction will have been fulfilled; if it breaks through, native lending will be serving demand other than financing shorts, and we will have to measure what that is.

Sources and links: Hyperliquid Documentation, portfolio margin (rate curve, caps, liquidations) · Hyperliquid Documentation, funding (interest component and formula) · Hyperliquid API info endpoint documentation (allBorrowLendReserveStates, metaAndAssetCtxs, fundingHistory) · HyperLend API, markets on HyperEVM · DefiLlama, HyperLend · DefiLlama, Felix · The Defiant, portfolio margin pre-alpha on mainnet (Dec-2025) · CoinDesk, move to alpha on Mar-10, 2026 · Hyperliquid on X, beta with expanded limits (Jun-25, 2026) · CoinGape, Hyperliquid post from Sep-18, 2026 on manual borrows · Cryptopolitan, opening of manual borrows