Aschenbrenner's AI Fund Cut Its Bitcoin Miner Book by 96%, From $1.83B to $80.8M
Leopold Aschenbrenner's Situational Awareness dumped IREN, CleanSpark, and Bitdeer hard in Q3 and wiped Riot, Core Scientific, and Hive off the sheet entirely. The real story isn't that the AI-power thesis broke. It's that a forced seller doesn't care about your thesis.
I keep a note on my phone for moments when the AI trade and the bitcoin mining trade stop being the same trade. Last Monday's 13F was one of them.
Situational Awareness, the AI hedge fund Leopold Aschenbrenner started in September 2024 after leaving OpenAI, reported about $80.8 million left across three miner stocks as of Sept. 30. Three months earlier it listed $1.83 billion across six.
That's a 96% cut. Not a trim. A haircut with a lawnmower.
The Receipts
Start with what stayed. The fund held 3,019,029 CleanSpark shares worth $38.9 million, 1,934,618 Bitdeer shares worth $20.3 million, and 527,864 IREN shares worth $21.6 million. Those are the only miner lines left in the filing.
Now the blood. IREN took the worst of it. Aschenbrenner's team held 9,474,099 shares worth $433.3 million at midyear. By Sept. 30 that was 527,864 shares. A 94% cut in share count.
CleanSpark went from 12,276,139 shares worth $178.6 million down to 3,019,029. Roughly a 75% trim. Bitdeer fell about 44%, from 3,439,450 shares worth $54.6 million to 1,934,618 shares worth $20.3 million.
And then there's the part that should make anyone holding miner bags sit up. Three names vanished completely. Core Scientific, 26,008,473 shares worth $665.6 million at June 30. Riot Platforms, 17,100,000 shares worth $468.2 million. Hive Digital, 9,133,726 shares worth $33.2 million. Poof. No line in the new report.
That's $1.167 billion in miner exposure that simply isn't there anymore.
Here's the mechanic most people miss. A 13F only reports long positions in U.S.-listed stocks and certain options. It doesn't show cash, it doesn't show shorts, and it doesn't show private holdings. Option lines get valued at the underlying shares, not the premium actually paid. So reading the headline drop from $20.24 billion across 26 lines to $1.58 billion across 23 lines as "the fund lost 92% of its money" is flat wrong. Reporting on a 13F like it's a bank statement is one of the most reliable errors in financial media.
The fund itself was blunt about the damage. In a letter to investors, Aschenbrenner said the portfolio fell an estimated 67% in July and that the fund had come "closer to permanent capital impairment than is acceptable." Year to date it was still up 80%. He said the fund wasn't shutting down and had kept its private investments.
What it looks like now is concentrated. The biggest common-stock line is SharonAI, 8,070,950 shares worth $440.7 million. There are call options on Bloom Energy, SanDisk, Nebius, and a memory ETF. The miners are a stub.
This Was a Forced Sale, Not a Thesis Change
So was Aschenbrenner wrong about electricity and compute being the scarce inputs to AGI? I don't think so. He's probably still right. He just got carried out on a stretcher before the thesis had time to pay.
Remember the July sequence. The fund sold the bulk of its public stock book to Ken Griffin's Citadel in a block trade after heavy AI losses. Reuters sources said it was forced to unwind most of its public holdings. In the letter, Aschenbrenner said he sold to remove all take advantage of and close shorts.
That's the sentence to reread. take advantage of and shorts.
When you're levered into a basket of miners and the AI complex cracks, you don't get to be picky. You sell what's liquid. IREN and CleanSpark have real depth and real volume, which is exactly why they got hit hardest on the way out. Core Scientific and Riot got liquidated to zero in the report because that's what a block trade does. It hands your whole position to somebody else in one afternoon.
And here's the thing about that Citadel deal. Somebody bought $1.1 billion of miners in late July. The smartest macro shop in the game looked at the exact same power contracts and data center leases and said yes. So the miners didn't get dumped because the assets are bad. They got dumped because one holder needed cash faster than the assets could deliver it.
That distinction matters for anyone trying to read signal into this filing. Fund flows aren't fundamentals. They're plumbing.
Meanwhile the fundamentals are genuinely mixed and nobody wants to say it out loud. IREN signed a multiyear cloud contract with Microsoft. CleanSpark locked in a $6.6 billion data-center lease. Those are enormous numbers. They're also future revenue against current capex, and miners have been burning cash on high-density buildouts while bitcoin trades below estimated production costs. JPMorgan flagged the unit economics getting worse this year. Great contracts don't fix a bad margin this quarter.
Big question nobody's answering: if power contracts are the crown jewels, why does the entire sector still trade on bitcoin's spot price?
What I'd Actually Watch
Here's my honest read. The AI-to-miner trade was a trade, not a religion. Anyone who bought CleanSpark or IREN because a hedge fund liked the power contracts needs to understand that the hedge fund has its own problems and its exit door is narrower than yours.
Two things to watch. First, mid-February. The fourth-quarter 13F will show whether those last three miner lines got trimmed too. If they did, the stub was just a leftover and the exit was total. If they didn't, you can read it as a floor.
Second, and this is the one that actually matters, the earnings reports. Watch whether the Microsoft contract and that $6.6 billion CleanSpark lease start showing up as contracted cash flow rather than a press release. If the revenue lands, the miners are power companies with a bitcoin hobby. If it doesn't, this whole sector was a proxy trade on someone else's AI conviction.
CT never misses. Except when it does.
The timeline is undefeated, and it just ate a $1.83 billion position alive in ninety days. Another day, another saga.