The chip stocks crashed at 5:22pm on a Saturday. Wall Street was closed
Stock perps trade all weekend on Hyperliquid. Last Saturday evening the ones tracking memory-chip makers fell six percent in ten minutes, with no exchange open anywhere to set a price. We went through every fill, forced close and price update, minute by minute.
- $43.8M
- 6.4%
- 82%
- $42.3M
- 1.1%
01 · Saturday, 5:22pm New YorkA crash with no exchange open
US stock exchanges close at 4pm on Friday. The Korea Exchange, where SK Hynix and Samsung are listed, closes at 3:30pm Seoul time the same day. Until Monday there is no price for either company anywhere, except on venues that never close.
Hyperliquid is one of those venues. Alongside its crypto perps it lists perps on single stocks, chip ETFs, gold and oil, run by a third-party deployer under the ticker prefix xyz. We have written about the exchange's paper ZEC market and its bridge before. On Saturday 17 October 2026 its stock markets did something no stock exchange can do: they crashed on a weekend.
The first fills that moved the SK Hynix perp arrived a minute after twenty past nine UTC, twenty past five in New York. Within a minute the forced closes began. A forced close, called a liquidation in crypto, is the exchange shutting a trade whose margin has run out, by selling it into the book. Each forced sale pushes the price lower, which runs the next trader out of margin. That is the chain reaction the chart shows.
02 · What was soldMemory chips, then everything near them
The markets that fell hardest make memory chips, or hold the companies that do. SK Hynix, Samsung and Micron are the three largest DRAM makers in the world. SanDisk makes flash. SOXL is a fund that moves three times as far as a basket of chip stocks. DRAM is a basket of the chip makers. Nebius, IREN and CoreWeave rent out the data centres the chips go into.
| Market, 21:00 to 22:00 UTC | Closed by force · accounts · longs |
|---|---|
| SK Hynix (xyz:SKHX) | $9.72M · 249 · 83% |
| Samsung (xyz:SMSN) | $4.41M · 101 · 84% |
| SOXL, 3x chip ETF | $3.99M · 184 · 76% |
| Micron (xyz:MU) | $3.88M · 179 · 84% |
| Gold | $3.21M · 4 · 98% |
| Crude oil | $3.01M · 2 · 18% |
| DRAM basket | $1.89M · 67 · 84% |
| SpaceX | $1.57M · 101 · 82% |
| Nebius | $1.42M · 145 · 94% |
| IREN | $1.12M · 105 · 99% |
Gold and oil look out of place on that list. They are there because of one account. The address 0x58f0bf4307c61bc7a5fe11e24fe36e64300b0d20 held Micron, gold and oil on shared margin at up to 25x. When Micron fell, the margin backing all three ran out together, and the exchange closed $7.4M of its positions in 620 fills. Four accounts made up the whole gold figure of $3.2M, and two the oil figure of $3.0M.
Most of the people caught were much smaller. The median forced close was a few hundred dollars at ten times margin, and the great majority were bets that the stocks would rise.
| Forced closes, 21:20 to 21:45 UTC | All stock perps |
|---|---|
| Positions closed by force | 17,606 across 53 markets |
| Accounts | 1,776 |
| Value closed | $43.8M |
| Longs | 82% |
| Median forced close | $407 |
| Median margin multiple | 10x |
| Closes at 20x or more | 907 |
| Realised loss on the closes | $2.4M |
03 · Book against oracleThe price that fell, and the price that did not
Every perp on Hyperliquid has two prices. The mark price comes from the order book, what people are bidding and offering right now. The oracle price is the exchange's view of what the asset is worth, fed from outside. Forced closes fire on the mark. When the underlying stock is closed, the oracle for an xyz market drifts slowly from its last reference, while the mark can go wherever the book takes it.
On Saturday the two came apart.
In twelve minutes the SK Hynix book fell more than four times as far as its oracle. Every forced close in those minutes was priced off the book.
| SK Hynix perp, 21:20 to 21:32 UTC | Move |
|---|---|
| Order book (mark price) | $914.18 → $855.88, down 6.4% |
| Oracle price | $916.54 → $902.96, down 1.5% |
| Book under oracle at the low | 5.2% |
| External reference, Friday's last print | $949.65, unchanged all weekend |
The same shape repeats across the chip markets. In each one the book fell two to four times as far as the oracle.
04 · Who soldA churner opened it, two accounts dumped Micron
The first aggressive seller in SK Hynix was a bot that traded both ways. It walked the price down over two minutes and bought almost as much back, the pattern of a market maker rather than someone getting out.
| First seller in SK Hynix | 0x1a3c983c2af04233bb26c77f44058a2d30b74e3d |
|---|---|
| First fill | 21:21:09 UTC |
| Sold in the first two minutes | $379,000 in 216 fills, $913.72 down to $901 |
| Bought back in the same two minutes | $309,000 |
| Over the crash, both SK Hynix markets | sold $4.9M, bought $2.5M |
Micron was different. Two accounts started selling in the same second, and between them they took the book down six percent in two minutes. The biggest forced close of the night, the gold-and-oil account above, followed.
| Micron, from 21:28:05 UTC | Aggressive sells |
|---|---|
| 0xab961d7c42bbcd454a54b342bd191a8f090219e6 | $8.2M in 161 fills |
| 0x0b7879fdafb34632b6b92287158bbb8705cf11f0 | $5.2M in 911 fills |
| Micron book, next two minutes | $874 → $823 |
Across all stock perps, the sellers outweighed the buyers by about twenty million dollars. In a book this thin on a Saturday, that was enough.
| All stock perps, 21:20 to 21:45 UTC | Taker side |
|---|---|
| Aggressive sells | $128M |
| Aggressive buys | $108M |
| Accounts crossing the spread | 3,213 |
| Fills | 213,332 |
05 · Ten minutes laterETH went next
The stock perps settled within a quarter of an hour. Then the forced closes started again, this time in ETH, and the worst single minute of the night was an ETH minute. ETH's book fell about a third as far as the chip stocks did, and two accounts at twenty-five times margin took most of the damage.
| ETH, Saturday 21:37 to 21:39 UTC | Closed by force |
|---|---|
| Minute of 21:38, all markets | $42.3M across 757 accounts |
| ETH market, full hour | $41.8M |
| ETH book | $2,420 → $2,354, down 2.8% |
| 0x30afce2f6842bf183c7e3fe7162e279ff0b6393e | $18.9M of ETH longs at 25x |
| 0x020ca66c30bec2c4fe3861a94e4db4a498a35872 | $12.8M of ETH longs at 25x |
Whether the stock crash caused the ETH drop or both answered the same selling, the record cannot say. It can say the order: stocks first, then ETH, ten minutes apart.
06 · The spot tokensOn Solana the same stocks fell one percent
Hyperliquid's xyz markets are paper. Nobody holds a share. The other way to trade a stock on a weekend is a token backed by a share in custody, and we have measured how those price in Monday's move and how xStocks trade on Solana.
Those tokens saw the same hour and shrugged.
| xStock on Solana, 21:00 to 22:00 UTC | Hour open → low |
|---|---|
| Nvidia (NVDAx) | $214.37 → $212.03, down 1.1% |
| Tesla (TSLAx) | $351.90 → $347.64, down 1.2% |
| S&P 500 (SPYx) | $764.92 → $758.73, down 0.8% |
A pool of tokens and dollars on a DEX has no margin and no forced closes. A perp book does. Same hour, same news, one market fell six times as far as the other.
07 · By MondayIt was as if nothing had happened
By Sunday evening, before US stock futures reopened, the SK Hynix perp sat less than one percent under where it had been before the crash. By Monday noon UTC, after the Korea Exchange had traded a full day, it was well above it.
| Market | Before crash → low → Sun 21:59 → Mon 12:00 UTC |
|---|---|
| SK Hynix | $914 → $855 → $906 (0.9% below) → $972 (6.3% above) |
| Samsung | $156 → $147 → $155 → $165 |
| Micron | $873 → $817 → $868 → $909 |
| SOXL | $93.2 → $86.7 → $92.2 → $103 |
| SanDisk | $1,293 → $1,215 → $1,280 → $1,354 |
The open interest in the stock perps did not change through the hour. What changed was who held it. About two thousand long positions were gone by the top of the next hour, and the shorts barely moved.
| All stock perps, hourly funding snapshot | 21:00 UTC → 22:00 UTC |
|---|---|
| Open interest, one side | about $1.84B → about $1.84B |
| Long positions | 71,066 → 68,966 (2,100 fewer) |
| Short positions | 22,528 → 22,377 |
08 · The scaleThe biggest hour these markets have had
In the thirty days we hold, no hour on Hyperliquid's stock perps comes close, and the next biggest came during US trading hours. Saturday as a whole beat the four weekend days before it put together many times over. Only one weekday, when US markets were open and selling chips all day, beat it.
| Stock-perp positions closed by force | Value · accounts |
|---|---|
| Sat 17 Oct, 21:00 hour | $43.9M · 1,794 |
| Next biggest hour, Thu 8 Oct 17:00 | $28.9M · 344 |
| Sat 3 Oct, whole day | $5.1M · 24 |
| Sun 4 Oct | $0.1M · 24 |
| Sat 10 Oct | $0.02M · 4 |
| Sun 11 Oct | $0.3M · 27 |
| Sat 17 Oct, whole day | $49.4M · 1,888 |
| Sun 18 Oct | $0.9M · 150 |
| Wed 14 Oct, a weekday, for scale | $105M · 3,891 |
09 · What people saidChecked against the record
The crash was all over crypto Twitter by Sunday. Coin Bureau, in a post about circuit breakers, wrote that $51.6 million of chip-stock bets "were liquidated there in just 10 MINUTES, per CoinGlass data." Our count for Hyperliquid alone is $43.8M over the full wave, with most of it inside ten minutes. CoinGlass counts more venues than one. The order of size is right.
Crypto Rover wrote that "major AI stock perpetuals plunged 5% within SECONDS, then quickly bounced." The fall took ten minutes and the bounce the next quarter of an hour. The size is right. The speed is not.
Ran Neuner wrote on Sunday that "tokenized AI stocks are already trading in the RED this weekend." Hyperliquid's stock perps sat about one percent under Friday by Sunday evening. The xStocks on Solana were down about the same. Deep red it was not.
10 · MethodWhat we counted
Bitquery indexes every fill, forced close, funding payment and price update on Hyperliquid. Forced closes come from the exchange's own liquidation fills, valued at the fill price. Taker volume counts only the side that crossed the spread, so no fill is counted twice. Mark and oracle prices are the exchange's own per-second updates, taken as minute lows. The xStocks prices are hourly candles from DEX trades on Solana. Everything here comes from the public record, and the MCP server we used to pull it answers the same questions in plain English.
Forced closes are valued at the exchange's liquidation fill price and summed over the window stated. They are not the traders' losses; realised loss on the stock-perp closes in the 25-minute window was $2.4M.
The xyz markets are run by a third-party deployer on Hyperliquid's HIP-3 standard. We describe their mark and oracle prices as the exchange publishes them and make no claim about how the oracle is built.
Addresses are described by what they did in the window. We do not know who holds them, and selling into a thin book on a weekend is not against any rule we are aware of.
Open interest is estimated from hourly funding records and rounded. The count of long and short positions is the number of accounts with a position in each direction, not its size.
Our trade table for Hyperliquid holds about thirty days, so the comparison with earlier hours and days stops at 19 September 2026.
This article is provided for informational and educational purposes only and reflects analysis of publicly available on-chain data as of the dates indicated. It does not constitute legal, financial, compliance, or investment advice, and nothing in it is a recommendation to buy, sell, or hold any token or asset.
The findings describe fills, forced closes and price updates recorded on Hyperliquid and trades recorded on Solana between 19 September and 19 October 2026. They may be incomplete or incorrect and may be revised as more data becomes available.
References to Hyperliquid, the xyz markets, xStocks, Coin Bureau, Crypto Rover, Ran Neuner, CoinGlass or any named company or person describe what the public record shows. They are not statements about any party's compliance, solvency, intent or conduct. Perpetual contracts on stocks are high-risk products that are restricted or unavailable in many jurisdictions.
Nothing herein should be relied upon as a definitive determination of fact. Readers should conduct their own independent verification before taking any action. The authors and publisher accept no liability for any loss or damage arising from reliance on this material. All trademarks and company names are the property of their respective owners.
Reported by Divyasshree for Bitquery Research, with AI tools; every figure was checked against the raw data.
Watch the next weekend crash as it happens
Every Hyperliquid fill, forced close and price update is in Bitquery's data within seconds. Stream it, query it, or ask the MCP server what the book did while the stock market slept.