Aschenbrenner's $13.67B fund bets $7.46B against chip stocks

Ex-OpenAI researcher Leopold Aschenbrenner has nearly doubled his Situational Awareness fund to $13.67 billion as of March 31, according to a Q1 2026 13F filed with the SEC on May 15 and accepted May 18. The filing reveals a concentrated bet: long positions in a basket of Bitcoin miners and energy infrastructure companies, offset by $7.46 billion in put options against the semiconductor sector. The pairing is internally consistent. Aschenbrenner has argued for months that the AI race will be bottlenecked by electricity and computing power, not chip design. If that thesis holds, chip valuations face a structural headwind even as power-site operators gain. For traders, the disclosure is a concrete map of one of the largest single-manager bets on that thesis in public equities.

Risk Event: $7.46B in chip sector puts

The fund's largest short-side positions are $2.04 billion against the VanEck Semiconductor ETF (SMH), $1.57 billion against Nvidia (NVDA), $1.07 billion against Oracle (ORCL), and $1.01 billion against Broadcom (AVGO). Combined, these four positions account for $5.69 billion of the $7.46 billion put book. The magnitude is rare for a single fund and signals a conviction that semiconductor valuations are disconnected from the real bottleneck in AI buildout.

Mechanism behind the thesis

Aschenbrenner's 165-page paper "Situational Awareness: The Decade Ahead" argued that compute infrastructure, not model development, would determine the pace of AGI progress. His investment logic follows directly: Bitcoin miners already own high-density power sites and grid interconnects that hyperscalers cannot replicate quickly. As miners repurpose those facilities into AI data centers, the value creation shifts from chip suppliers to site operators. The puts against SMH, NVDA, ORCL, and AVGO are a hedge–or a direct bet–that the market is mispricing where the infrastructure value chain goes.

Exposure: Miners as the long side

The fund's largest disclosed long positions are all Bitcoin miners: IREN (IREN), Core Scientific, Riot Platforms (RIOT), CleanSpark, Bitfarms, Bitdeer, and Hive Digital. These are joined by energy and compute plays Bloom Energy, SanDisk, and CoreWeave. The rationale: miners have the power capacity, land, and cooling infrastructure that AI data centers need. Core Scientific has announced plans to repurpose its Pecos site into a 1.5GW AI data center campus, converting 300MW of existing mining capacity. TeraWulf reported that AI and HPC hosting revenue of $21 million outpaced Bitcoin mining revenue for the first time in Q1 2026.

Proprietary AlphaScala data

At current prices, NVDA carries an AlphaScore of 67 (Moderate), RIOT scores 80 (Strong), and IREN scores 34 (Weak) on AlphaScala's proprietary model. The divergence reflects the market's early-stage pricing of the miner-to-AI pivot. NVDA stock page | RIOT stock page | IREN stock page

Timeline and catalyst pace

The 13F covers positions as of March 31, 2026. The fund doubled its disclosed equity exposure from $5.52 billion at end-2025, indicating aggressive deployment in Q1. The puts were likely opened or increased during the same period. The next meaningful catalyst is Q2 2026 filings due mid-August, which will show whether the fund added to or trimmed the chip puts as semiconductor stocks moved in Q2. The miner AI revenue reports in the current earnings season (TeraWulf already showed the inflection) will test the thesis on the long side.

What would reduce the risk for chip stocks

• A clear signal that chip shortages persist despite power constraints. If hyperscalers accelerate chip orders and Nvidia's guidance beats, the thesis that power is the bottleneck weakens.
• Miner AI conversions fail to deliver margins. If Core Scientific's Pecos project sees cost overruns or lower-than-expected utilization, the miner long side loses credibility and the chip puts could be unwound.
• Regulatory or grid-access changes that make new data center builds cheaper, undercutting the value of existing miner sites.

What would make it worse for chip stocks

• More miner earnings showing AI hosting margins beating mining margins. A second quarter of TeraWulf-style data would validate the power-as-bottleneck thesis and could spark revaluation of the entire miner sector.
• A surprise buildout delay from a major chip customer (e.g., Microsoft or Meta cutting data center capex). That would compress chip orders and align with Aschenbrenner's short thesis.
• Follow-on 13F filings from other large funds showing similar miner-long/chip-short positioning. A crowded trade would amplify moves.

Affected assets summary

The broader miner pivot

Aschenbrenner's bet is one expression of a wider sector shift. Crypto.news documented how Bitdeer, Riot, and others are accelerating the conversion of mining facilities into AI data centers. The fund's 13F is the largest disclosed single investor conviction on that trade. The question for traders is whether the chip puts are a hedged overlay or a directional short. The magnitude relative to the miner longs suggests a directional view: the fund is net short semiconductors by a wide margin, betting that the market will soon reprice where AI infrastructure value actually resides.

For now, the trade is early stage. The puts have time value and the miner longs are still priced largely as crypto plays. If the thesis earns confirmations from more earnings reports, the NVDA and SMH shorts will face tailwinds. If not, the fund may need to roll or close the puts. Either way, the 13F gives traders a rare, timely map of a concentrated institutional bet on the AI power bottleneck.