Aschenbrenner's Situational Awareness fund cut its stake in bitcoin miner IREN by 94% in its Q3 filing.
**Miner Stakes Slashed in Q3 Filing**
Leopold Aschenbrenner, the former OpenAI researcher behind the hedge fund Situational Awareness, cut his Bitcoin (BTC) mining stakes sharply in the third quarter, according to the fund's latest quarterly 13F filing with the U.S. Securities and Exchange Commission (SEC). The filing shows a 94% reduction in IREN Limited (IREN), a 75% cut to CleanSpark (CLSK) and a 44% cut to Bitdeer (BTDR) compared with the June quarter. The fund also fully exited Core Scientific (CORZ), Riot Platforms (RIOT) and Hive Digital (HIVE). Those three exits eliminated the fund's entire disclosed crypto mining exposure.
The positions had been built on a simple overlap. Miners own grid-connected sites and long-dated power contracts, which makes them candidates to supply electricity for AI data centers as well as to produce bitcoin. Aschenbrenner's 2024 essay, *Situational Awareness: The Decade Ahead*, argued that spending on chips, data centers and power would run into the trillions of dollars, with electricity rather than chip output acting as the binding constraint. The Q3 report shows the mining side of that trade is now largely gone. The fund kept its stake in SharonAI, an Australian AI cloud provider, and moved some single-stock bets from common shares into listed call options.
Call options give the holder the right, but not the obligation, to buy shares at a fixed price, a structure that caps the cost of the bet at the premium paid. IREN, CleanSpark and Bitdeer remain on the register at reduced size, so the sector retains a foothold in one of the most closely watched accounts in the AI trade.
**From $20.24 Billion to $1.58 Billion**
The same filing explains why the mining book shrank. Situational Awareness reported $1.58 billion in US-listed stocks and options as of September 30, down from $20.24 billion in its June disclosure, a contraction of 92%. The collapse traces to July, when the fund's leveraged bets on technology shares moved against it and lenders demanded cash. The fund lost roughly 67% of its public book in that episode, and the surviving holdings were sold to Ken Griffin's Citadel.
What re-entered the filing is mostly upside exposure. About $1.04 billion of the September total sits in call options on Bloom Energy, Nebius, SanDisk and a memory-chip fund. These positions pay off only if the underlying stocks clear a set price before the option expiry date. The June report held those three names as common stock; the September report carries them as options instead. The choice keeps the fund within its own power-compute-memory framework: Bloom supplies fuel cells for data center power, Nebius sells cloud computing capacity and SanDisk makes flash memory.
Two limits on the disclosure matter for anyone reading the numbers. A 13F excludes private-company stakes and short positions, and it values options by the shares they cover rather than by the smaller premium paid. The $1.58 billion is therefore neither the fund's total assets nor a direct measure of its losses. The filing also does not show whether the mining cuts reflected a deliberate pivot or the mechanics of unwinding after a forced sale, because it records only position sizes at quarter-end.
**SharonAI Carries the Remaining Bet**
SharonAI now carries the fund's largest stock position. The filing lists 8.07 million shares, valued at $441 million on September 30, and that stake has aged badly. SharonAI (SHAZ) has fallen more than 60% since the filing date and was quoted at $20.48 in pre-market trading on Tuesday. Nebius has also dropped more than 60% over the same span. Declines of that size amount to a bear market for the two names at the core of the rebuilt portfolio.
The September 30 valuation predates the slide, so the current worth of the SharonAI stake is materially lower than the figure on the filing's face. Pre-market quotes are thin and can move sharply once regular trading opens, so the $20.48 price is an indication rather than a settled level.
Aschenbrenner founded the fund in 2024 around a single thesis: that artificial general intelligence by 2027 was plausible, and that the bottlenecks would be chips, data centers and power. The filing's other limits cut both ways. It does not capture the fund's roughly $10 billion of total assets, most of which sit in private stakes such as Anthropic, and it cannot show any trading done in October. The next 13F, due by mid-February, will show whether Aschenbrenner held the remaining stock positions through the slide or cut again. Until then, the public record is a September snapshot: a fund that once disclosed a $20 billion public book now reports $1.58 billion, and its largest single stock has lost most of its value since that picture was taken.
**A Buyer Steps Away From Bitcoin Miners**
Read together, the disclosures say less about the mining-to-AI conversion trade than about what leverage did to one prominent buyer. Aschenbrenner's thesis on power scarcity is unchanged, and the options book keeps the fund exposed to the same electricity, compute and memory complex. What changed is the vehicle. The miners are out of the common-stock register, Core Scientific, Riot and Hive have lost a visible institutional holder, and the sector's best-known advocates no longer include this fund.
IREN, CleanSpark and Bitdeer remain on the register in reduced form, so the conversion story keeps a foothold in a closely watched account. With SharonAI quoted at $20.48 before Tuesday's open and no new filing due before mid-February, there is no public record yet of whether the remaining positions have moved further.