Why a $108.9M long ETH 10× on Hyperliquid? Reconstruction, microstructure, revealed intention

The position is a long ETH-USD perpetual swap: 41,230 ETH at 10x cross leverage, entry price $2,693.92, mark $2,641.50, break-even $2,698.88. Notional is $108.91 million, with unrealized PnL of -$2.16 million and cumulative funding of -$137,658. Liquidation price is $2,514.72, implying 4.80% from mark and 6.65% from entry. Implicit equity is $8.81 million (the $7.35 million UI cash shows separate spot wallet). Effective leverage is 12.4x (vs 10x initial). Hyperliquid’s liquidation formula: S_L = S – side m_avail / [q (1 – ℓ side)], where side=+1 (long), ℓ=1/L_maint, and in cross m_avail = E – MM. ETH’s margin tier: up to $100M notional → max leverage 25x (MM ~2.0%); above $100M → max 15x (MM ~3.33%). This position sits at the $100M boundary. In cross, displayed 10x only sets initial margin at opening: IM = $11.11 million. Current equity = IM + uPnL + funding: $11.11 – $2.16 – $0.14 = $8.81M. Consistency check: loss to S_L = 41,230 ETH ($2,641.50 – $2,514.72) ≈ $5.23M; equity at S_L ≈ $8.81M – $5.23M = $3.58M.

Notional at S_L ≈ $103.7M; implied MM rate ≈ 3.45%, aligning with >$100M tier (~3.33%) plus fee/funding/tier deduction. The $7.35M overview shows only cash spot, not total equity. Distance to liquidation as return: 4.80%. With ETH 30-day realized vol 45-70% annual, a 1-day move of 4.8% is about 1.0-1.4 standard deviations (assuming σ=50% annual, σ_1day~3.1%), making it an ordinary risky day, not an extreme tail event.

On September 21, an address held ~4,487 ETH long, then bought 24,738 ETH (~$66.6M) at ~$2,693.21 in 21 minutes, increasing holdings to ~29,225 ETH (6.5x) with a liquidation price near $2,394. This was a continuation scale-in above the prior cost basis during a rally, not a distressed average-down. By late on September 21-22, the position reached 41,228.48 ETH, average entry fixed at $2,693.92, 10x leverage, making it the 4th largest ETH long on HL with a notional of ~$111-112M. Eleven hours later, unrealized PnL was +$1.524M with ETH at ~$2,734. The trader’s history includes two closed BTC and ETH longs with +$3.283M realized profit and a 100% win rate, though the sample is too small for statistical edge. By September 24, despite the position swinging from +$1.52M to -$2.16M (a -$3.68M mark-to-market change over ~48-60 hours) and $138k in funding costs, the trader neither cut nor added, showing swing trend-follower behavior holding the signal, not a scalper or market maker.

The long 10× position is not a market-maker trade, as it lacks a visible hedge or bid-ask inventory. It is not a basis or cash-and-carry arbitrage, since the spot UI is in cash, not physical ETH. Insider ETF activity is unlikely because the ETF creations on September 21–23 were already public and priced in. The 10× leverage is lower than the average for major ETH holders (12–13×), far below the 25× of a casino gamble, and conservative relative to the 3n long at 25× with a -3.4% stop-loss.

The sizing is aggressive in notional but disciplined in leverage. The central hypothesis is that this is an institutional-compatible momentum trade executed on-chain. Between September 21–23, the market saw $999M and $715M inflows into BTC ETFs, and $270M and $162M into ETH ETFs, with BTC testing $86–87k and ETH following as beta. An agent trading flow convexity (ETF creation = demand for the underlying, delta-hedging by APs, short squeezes) would choose ETH over BTC if they had already monetized a BTC long (with $3.28M realized profits), if ETH offers a beta >1 on a risk rebound, and if the Hyperliquid order book (with ~$2.9B in open interest) can absorb 40k ETH without excessive slippage when entered over 21 minutes.

The 10× leverage is a standard compromise for a prop desk or crypto whale: enough to generate 50–80% ROE from a 5–8% trend, but with sufficient cushion to survive a 4–6% pullback, close to the current stop-loss distance. This is empirical fractional Kelly, not quantum optimization. The implicit objective function over a 3–10 session horizon considers expected return μ_T (rebound post-ETF or hold above 2,600), volatility σ_T, carry cost c_T (hourly funding of 0.00125%, ~11% annualized as of September 24), and ruin cost Ψ = Pr(S_T ≤ S_L)·E. The program is to maximize leverage L(μ_T – c_T)E_0 – λΨ(L) – κ·Impact(q).

At 25×, Ψ explodes with a stop-loss 3–4% below spot (like the 3n long). At 3–5×, ROE on a bounce from 2,700 to 2,800 is too low to justify $11M in locked margin. 10× is the interior where payoff-to-ruin distance is acceptable for an agent who accepts a daily liquidation probability of 10–20% under a naïve lognormal assumption, high for a UCITS fund, routine for a crypto perp desk. Hyperliquid is chosen over a CEX due to voluntary on-chain disclosure (the address was public and picked up by Ai Yi in <12 hours). This signals either indifference to publicity (branding/confidence) or a constraint: size, speed, no KYC, hourly funding, no position limits. The cost is $138k in funding already, plus the risk of being a cascade target, with the stop-loss clearly visible at $2,514.72.

Microstructure Analysis of ETH Position: This position is analyzed as a market object. ETH HL Open Interest is approximately $2.87 billion (as of 24/09). Longs within the 3-5% range below the spot price total about $226 million. The analyzed position (~$109 million, -4.8%) is a key component of this range. A print at $2,515 is not an isolated whale liquidation but an ~0.4% supply shock to ETH HL OI, affecting neighboring range positions and potentially triggering ADL/HLP if the order book fails. A lower long cluster is near $2,443 (~$141 million). The largest long (8x, ~$150 million, stop-loss ~$2,446, -7.5%) is further out. The largest short (~$206-278 million, 5x, stop-loss > $4,300) is not a hedge for this wallet but a separate agent far from its stop. Hourly funding rate ~+0.00125% (longs pay, ~11% annualized). Carry cost on $109 million: ~$1,360/hour or $32.6k/day. Total carry of $137.7k / $1,360 per hour = ~101 hours ≈ 4.2 days, consistent with a position opening on the evening of the 21st if rates were slightly higher during the 22-23 squeeze. Each day holding without a bounce to break-even price ($2,698.88) creates option-like depreciation from carry, not just mark-to-market. Impact: 41k ETH equals $109 million. On a deep perpetual order book, the entry was executed sharply (at least one 21-minute block or $66 million). Exiting in a down-tape near $2,520 would incur significant slippage, incentivizing holding or adding collateral rather than market-selling 41k ETH into the cascade the address itself would trigger.

A risk committee (UCITS or SEC/CFTC-regulated desk) would not permit 12x leverage on a single crypto perpetual with 4.8% distance-to-ruin and 100% equity concentration. The 99% 1-day VaR under 50% volatility appears benign at ~$0.64M, but this metric is flawed: losses are non-Gaussian and absorbed at a -4.8% barrier. The correct measure is a discrete Merton-style distance-to-default: DD = ln(S/S_L)/(σ√τ). For 1-day, 50% annual vol: ln(2641.5/2514.7)/(0.50/√365) ≈ 1.88, implying 3.0% default probability under lognormal assumptions. Under Student-t with 4 degrees of freedom (typical for crypto fat tails), the barrier is significantly higher.

A historical stress event, a 5% ETH down-day (frequent in 2022-2024), would already liquidate the book. Tax treatment (general, not advice): crypto perps are typically treated as financial instrument P&L or capital gains depending on residence (US: mark-to-market possible for 475 traders; France: capital gains regime for digital assets is ill-suited to perps, often BNC/capital gains; Switzerland: wealth tax plus potential professional activity status). Trading without KYC does not eliminate reporting obligations if the agent is an identifiable taxable person. Funding paid is an expense, not a disposal capital loss.

---

1. The agent is a whale-scale trend-follower: it scaled ×6.5 during the rally on 09/21, not at a trough. 2. The 10× leverage is a medium-high conviction bet over a few sessions, not a 25× all-in; the notional is the real risk, not the label. 3. The choice of ETH over BTC after already cashing in BTC/ETH longs (+3.3M$) is a beta rotation in an ETF regime that remains bullish. 4. It did not cut at +1.5M$ or -2.2M$: the stop rule is not a PnL tick, but likely a structural break (loss of 2,500–2,520 / death of ETF flow). 5. As of 09/24 1:00 PM CEST, the position is underwater, 4.8% from the barrier, with negative carry. The “infallible genius” hypothesis is already falsified on a mark-to-market basis. The hypothesis that “he is right over 5–10 days if rates stop rising” remains open. 6. For the rest of the market, S_L = 2,515$ is a liquidity magnet: either it is defended by additional equity, or it becomes a forced offer.