On Friday, 14 August, a firm called Situational Awareness LP filed its Form 13F with the Securities and Exchange Commission. The document describes an equity portfolio of $20.24 billion spread across 26 positions. It is an ordinary filing, correctly completed and submitted on time.
The portfolio it describes no longer existed on the day it was filed. It had been sold two weeks earlier, under duress, to a single buyer.
There is no sharper illustration of what a 13F is and what it is not. Anyone who wants to understand why copying famous portfolios from these filings does not work need not read a study. It is enough to set this one form next to a calendar.
What the document says
As of the 30 June 2026 reporting date, Situational Awareness reports a portfolio worth $20.24 billion in 26 positions. Two of them account for most of it. SanDisk, at 2,495,344 shares and $5.67 billion, represents 28.03 percent of the book. Micron Technology, at 4,828,786 shares and $5.57 billion, represents 27.54 percent. Together that is 55.57 percent of the reported portfolio in two makers of memory chips.
Behind them come Bloom Energy at 9.38 percent, Taiwan Semiconductor at 6.25 percent and the newly added, Amsterdam-founded Nebius Group at 6.09 percent. The remainder is spread across data-centre operators, bitcoin miners and power names: CoreWeave, Core Scientific, Applied Digital, Riot Platforms, IREN, CleanSpark, Bitdeer. It is a portfolio built entirely on one thesis, and the thesis is that the build-out of computing capacity for artificial intelligence continues and that the bottleneck sits in memory.
How new that thesis was becomes clear only against the prior quarter. On 31 March, Micron sat in the book at 17,362 shares. By 30 June it was 4,828,786 — an increase of 27,712 percent, which is to say a position built from nothing. Taiwan Semiconductor went from 22,423 to 2,649,035 shares, up 11,714 percent. SanDisk roughly doubled, from 1,140,119 to 2,495,344 shares. In a single quarter, more than eleven billion dollars was directed into two securities that had been all but absent three months earlier.
The reversal that only the comparison reveals
The most revealing part of this filing is not in it. It emerges from the difference to the previous quarter. Twenty-one positions reported on 31 March are entirely absent on 30 June — and most of them were put options.
In the first quarter the fund held puts on a semiconductor index fund worth $2.04 billion, on Nvidia worth $1.57 billion, on Oracle worth $1.07 billion, on Broadcom worth $1.01 billion, on AMD worth $969 million, on Micron itself worth $584 million, on Taiwan Semiconductor worth $535 million, on ASML worth $494 million and on Intel worth $159 million. Added up, that is more than eight billion dollars of puts on precisely the chain whose shares the fund would concentrate into one quarter later.
Placed side by side, the two filings do not show an adjustment. They show a complete reversal of direction inside three months: from a broadly hedged — or outright bearish — stance on AI semiconductors to an unhedged concentration in two memory manufacturers.
One qualification is needed here, and it is the one most often missed with options in a 13F. The value shown for a put position is not the money at risk; it is the market value of the underlying shares. A $1.57 billion put on Nvidia means the options reference $1.57 billion worth of Nvidia stock. What was actually paid is the option premium, which depending on strike and maturity is typically a low single-digit percentage of that. Strike prices and expiry dates do not appear in the form at all. How large the hedge really was in economic terms therefore cannot be determined from a 13F. That it vanished completely in the second quarter can be.
Three things missing from the form that decided everything
At the reporting date the $20.24 billion book looked concentrated, but not fatal. What it actually was becomes visible only through three facts a 13F structurally cannot contain.
First, leverage. A 13F reports positions, not equity. It shows what a manager holds, never how much of it is borrowed. According to consistent reporting, Situational Awareness ran gross leverage of around 400 percent. The same $5.57 billion Micron line means, at an unlevered fund, that $5.57 billion of the manager’s own capital sits behind it; at four times leverage it means roughly $1.4 billion does and the rest comes from prime brokers. In the filing the two cases are indistinguishable. This is the single most consequential omission in the entire disclosure regime, and it appears nowhere as a footnote.
Second, anything not listed in America. The reporting obligation covers US-listed securities. SK Hynix was reportedly among the fund’s core positions. SK Hynix trades in Seoul and therefore appears in no 13F, this one or any other. Anyone tracking this firm’s memory bet through the American form saw part of it without knowing how large the missing part was. It is the same mechanism that keeps Berkshire’s stakes in the five Japanese trading houses invisible.
Third, anything not publicly traded. The fund held a stake in Anthropic valued at roughly five billion dollars. Private holdings are not reportable. They appear in no line, no total, no weighting. As it turned out, that was precisely the part of the balance sheet that survived the following weeks.
What happened between the reporting date and the filing
The deadline for a 13F is forty-five days after quarter end. The reporting date was 30 June; the filing landed on 14 August. Here is what those six weeks contained.
Through the end of June the fund had reportedly returned 439 percent, and assets under management peaked in early July at around $45 billion. Then competitor SK Hynix reported a record quarter, with operating profit up 557 percent year on year, and in the same breath guided 2026 capital expenditure fifty percent higher, to at least $31 billion. A record result thereby became a sell signal: if the most profitable supplier is adding capacity that aggressively, the scarcity holding prices up is ending. The market read it as the beginning of a glut.
Prices followed. Micron fell about twenty percent in July, SanDisk more than thirty-five. Korea’s benchmark index dropped 29 percent in a month. The fund’s specific positions reportedly fell between 35 and 47 percent.
At four times leverage, that is sufficient. A 35 percent decline on the gross book wipes out the equity supporting it. The fund’s prime brokers — among them Bank of America, Goldman Sachs and JPMorgan — called for margin. When it did not arrive in sufficient size, the book was sold: Citadel absorbed roughly sixteen billion dollars of positions at a discount of about ten percent, a process completed on 30 July. July closed with a loss of roughly 67 percent. The Anthropic stake, which never appeared in any 13F, remained.
Fifteen days later, the form describing that portfolio was published.
What follows from this, and what does not
The obvious lesson is that 13F filings are worthless. That is too easy, and in this form wrong.
The forty-five-day lag does not matter equally for every manager. It matters in proportion to turnover. At Berkshire Hathaway, where fifteen of twenty-nine positions did not move by a single share last quarter, a six-week-old filing still describes the present quite accurately. At a fund that turns 17,362 shares into 4.83 million within a quarter, it describes a state that was already history when the envelope was sealed. Same deadline, entirely different information content.
Nor is the opposite claim true — that copying successful managers is hopeless in general. There is research finding that portfolios assembled from the holdings of consistently successful managers still earn excess returns even after the disclosure lag. It is simply that no copier of this particular book would have been destroyed by the lag. The decisive variable was not in the form. Someone who had bought all 26 positions in the same proportions with their own money would have been down perhaps thirty to forty percent in July. Painful, survivable. The fund lost 67 percent and its book because four times the capital sat on the same positions. The line item was identical. The structure underneath it was not.
Which means the practical question to ask of any such filing is not what someone owns, but how concentrated the ownership is and what could force them to sell. On the first half the form answers well: 55.57 percent in two securities is stated plainly. On the second it says nothing at all.
Why no such filing exists in Germany or Austria
Investors in the German-speaking market who follow the American disclosure cycle often assume a European equivalent. There is none, and the difference is systematic. The United States regulates disclosure by the size of the manager: cross the $100 million threshold and your entire reportable equity book becomes public, down to positions worth a few hundred thousand dollars. Europe regulates by the significance of the stake: disclosure is triggered when a voting-rights threshold is crossed.
In Germany, notification duties under Section 33 of the Securities Trading Act begin at three percent of voting rights, with further steps at 5, 10, 15, 20, 25, 30, 50 and 75 percent, published via BaFin. In Austria, Section 130 of the Stock Exchange Act sets the first threshold at four percent, with steps up to 90 percent, supervised by the FMA. Europe therefore discloses faster, but only where influence is at stake. America discloses completely, but six weeks late.
For the case described here, none of that would have helped. A fund’s leverage is not public on either side of the Atlantic. No European regime would have made this structure visible either.
The number that mattered
Two figures from this filing deserve to stand next to each other. The first is 55.57 percent, the share of two memory manufacturers in the reported book. The second is 400 percent, the estimated gross leverage. The first is in the filing; every screener repeats it; it was still verifiably accurate on 14 August. The second appears nowhere.
It was the second that cost the fund. The first alone would have produced a bad quarter. Together they produced a forced liquidation on a Thursday in late July — sixteen billion dollars to a single buyer at a ten percent discount.
Reading 13F filings for ideas is a reasonable use of them. Reading them to reconstruct portfolios means working with a document that shows the positions and withholds the structural load they sit on. This quarter, the difference between the two is documented more clearly than usual.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
