Leopold Aschenbrenner, the former OpenAI safety researcher who left the company in April 2023, filed regulatory disclosures showing his hedge fund holds significant short positions against Nvidia and a basket of AI semiconductor manufacturers. The filings arrived within days of a separate billionaire manager liquidating a $47 million MicroStrategy stake in its entirety. Both moves mark rare public dissent against the AI infrastructure trade that added $2.1 trillion in market capitalization to semiconductor and proxy-exposure equities since January 2023.
Aschenbrenner's fund, launched in late 2024 with backing from several West Coast family offices, disclosed put options and direct short positions against Nvidia, Broadcom, and three Taiwan-based foundry suppliers. The positions were established between December 2024 and February 2025, according to the 13F filing. The billionaire manager's MicroStrategy exit occurred in a single block trade on March 14, liquidating the entire position accumulated over six quarters. MicroStrategy shares trade at 8.7x book value, a multiple sustained almost exclusively by the company's 214,246 Bitcoin treasury holdings. Nvidia closed Friday at $142.33, up 94% over twelve months but down 11% from its January peak.
The timing matters because both managers are making explicit second-derivative bets. Aschenbrenner is not arguing AI development will slow. He is arguing that current chip demand reflects pull-forward orders from hyperscalers racing to secure capacity, not sustainable run-rate consumption. His public writing since leaving OpenAI has focused on AGI timelines and compute scaling laws—he understands the technology. That makes the short position a valuation call, not a skepticism call. The MicroStrategy exit is simpler arithmetic. With Bitcoin at $67,400 and the company's treasury worth roughly $14.4 billion, the equity market is pricing MicroStrategy's operating business and future issuance optionality at $11 billion. That assumes continued access to convertible bond markets at favorable terms and no regulatory pressure on leveraged Bitcoin accumulation strategies. Both assumptions held through 2024. Neither is certain through 2026.
Allocators should note three follow-on pressure points. First, Nvidia's April earnings call will need to address data center utilization rates, not just order backlog. Hyperscalers have $190 billion in committed AI infrastructure spend through 2025, but actual workload deployment lags capacity installation by nine to fourteen months across AWS, Azure, and Google Cloud. Second, MicroStrategy's convertible bond maturities begin in March 2027 with $1.05 billion due. The company will need to either refinance or liquidate Bitcoin to meet that obligation if equity remains inaccessible. Third, the台湾 foundry suppliers in Aschenbrenner's short basket face margin compression as TSMC's Arizona and Germany fabs come online, shifting pricing power back toward customers.
The cleanest read is that sophisticated capital is rotating out of the second-order AI beneficiaries—the treasury plays and the capacity providers—while maintaining exposure to the actual model builders and application-layer companies. That rotation shows up in fund flows: AI-focused ETFs saw $830 million in outflows during February while software and SaaS funds absorbed $1.2 billion in new capital. The short positions are not market-timing. They are spread trades against crowded consensus.