Invest16 publishers2 min readPublished Updated
A 3% retrace cleared $475M in longs: the leverage was the news, not the price
Bitcoin slipped from about $79,500 to $77,000 on August 22 and roughly $547 million of crypto positions were force-closed. At 50x, a 2 percent move is the entire margin.
The Investor · Invest desk

What happened
- Bitcoin ran nearly 30 percent over five days and briefly touched about $79,500 before reversing on August 22, 2026.
- The retrace to around $77,000 force-closed more than $475 million of long positions in a short window.
- Total crypto liquidations for the episode came to roughly $547 million.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- cost Every percentage point of the pullback cost about $174 million in closed positions, and the bill landed on whoever bought the final leg of the run rather than on the venues that financed it.
- constraint While 50x and 100x remain on offer, a book of that gearing cannot absorb an ordinary day's range without generating sell orders, so position size rather than market direction sets the ceiling on...
- exposure Anyone who added into the top is now underwritten by a single level: if $77,000 does not hold, the same mechanism has a second tranche of margin to take.
- precedent Both sides of this market have been flushed inside one month, which means the next macro headline gets traded by participants who now expect a cascade in whichever direction they are wrong.
On a 50x perpetual, a 2 percent adverse move consumes the posted margin. At 100x, 1 percent does it. The distance from $79,500 down to $77,000 is 3.1 percent [1], which is more than one and a half times the wipeout threshold for a 50x long and about three times it for a 100x long [8]. Both ratios remain widely available on perp venues [12], so the forced selling was not an accident of thin books. It was the margin engine doing the only thing it does.
Of the roughly $547 million cleared, about 87 percent sat on the long side [2], consistent with derivatives data from CoinGlass showing the bulk of liquidations hitting bullish bets [7]. Hyperliquid saw individual liquidation events of significant size during the session, according to the same account [8].
The comparison worth making is with the other direction. Earlier in the month, the rise off $64,000 to $65,000 generated between $1 billion and $3.5 billion of short liquidations across various 24-hour windows [9]. The long flush is therefore somewhere between 14 and 48 percent of what the short flush cost [4]. The crowded side that everyone identifies after the fact was the smaller of the two positions the same book carried in four weeks.
The published figures also do not quite close. A near-30 percent five-day gain into $79,500 [2] implies a starting point near $61,150 [5], which is $2,850 to $3,850 below the $64,000 to $65,000 base given for the squeeze [6]. Either the five-day window opens before that base or the 30 percent is generous. Which figure you keep determines how much of the run you credit to short covering rather than to the Treasury's roughly doubled long-term bond buyback operations [10] and the administration's signals on clearer rules for crypto exchanges [11].
None of those macro inputs changed on August 22. Buying paused, and geared positions had nothing to lean on, with overbought readings at their most extreme since November 2024 [3]. At $77,400 to $77,500 [13], bitcoin is still about 20 percent above the base it left earlier in August [7]. The trend did not break; the financing did. Observers who describe events like this as a healthy reset of excess leverage [16] are naming a fee, not a repair.
What to watch
- Whether open interest rebuilds at the same leverage within days, which would tell you the flush changed positioning discipline not at all.
- Whether the Treasury keeps buyback operations at roughly double size, since it is the cited macro leg with an actual schedule.
- Whether any venue trims 100x access after a two-way cascade in a single month, or simply prices the volatility into funding.