Half a book in two memory names: what Situational Awareness's 67% month actually shows

Situational Awareness LP filed its 13F-HR with the SEC on August 14, reporting holdings as of June 30, and SanDisk plus Micron made up roughly 50% of the reported US equity assets ^[1]. Weeks later the fund told investors it had booked an unaudited 67% loss in July after Leopold Aschenbrenner sold down part of the public portfolio to reduce leverage once liquidity was exhausted ^[2][3]; Reuters confirmed both the July loss figure and the disposal of the bulk of the public book ^[4].

The distinction that matters for anyone sizing memory exposure is that the underlying demand signal did not break in July. Counterpoint Research put enterprise SSDs at 48% of total NAND shipments in Q2 2026, attributing the shift to AI workloads moving from training to inference ^[5]. TrendForce expects NAND Flash revenue to rise 10% to 15% quarter on quarter in Q3 2026 and DRAM prices to rise 13% to 18% ^[6]. Micron reported a record fiscal third quarter of $41.46 billion against $9.30 billion a year earlier, a figure the report describes as earnings and which is worth treating as reported rather than settled ^[7]. None of that is a thesis unravelling.

What did unravel was the structure around the thesis. The prior filing, for the quarter ended March 31, showed $5.52 billion in US equities alongside $8.7 billion in put options on chip companies, with Bloom Energy the largest long at $879 million, followed by SanDisk and CoreWeave ^[8][9]. That puts the options position at about 1.6 times the long equity book ^[1]. Total 13F securities came to $13.68 billion, which the reporting is careful to note is not the fund's total assets or its economic exposure ^[10]. Nor do 13Fs capture cash, shorts or personal holdings ^[11], which is why an anonymous X account's July 28 estimate of a $600 million loss across Bloom Energy and SanDisk went unverified by the publisher that surfaced it ^[12].

The return path tells you how much was riding on it. The fund reported the 67% July drop while still showing a year-to-date return near 80% ^[2]. If both figures hold, the book was up roughly 445% before July began ^[2]. A drawdown of that shape is what leverage plus two-name concentration produces when the financing side tightens: correct on direction, forced to sell anyway. Aschenbrenner's own framing to investors was that the fund "embraces volatility. But it should never jeopardize the fund," and that it had acted "to fight another day" ^[13].

He has not stepped back from the sector. His firm put $400 million into Source Foundry, a stealth semiconductor-equipment startup valued near $5 billion, working on lithography, the bottleneck ASML dominates ^[14][15]. The macro case remains large on paper: Goldman Sachs sees close to $7.6 trillion spent on compute, data centers and power between 2026 and 2031 ^[16], while the Semiconductor Industry Association and Deloitte put semiconductors at more than 95% of the value of an AI server rack and data-center infrastructure investment at up to $4 trillion by 2028 ^[17][18].

Watch three things. First, whether the audited July numbers match the unaudited 67% ^[2]. Second, the next 13F, which will show whether memory remains half the book or the concentration has been dialled down after the forced sales ^[1][3]. Third, whether TrendForce's Q3 price increases actually land ^[6], because a cyclical pricing trade financed with leverage is a different instrument from the same trade held unlevered.