Dan Loeb's Third Point LLC filed its Q2 13F showing a full exit from gold exposure and fresh stakes in former Bitcoin miners pivoting to AI datacenter operations. The rotation comes as Bitcoin logged its third consecutive winning week while equity and currency markets convulsed. Third Point did not disclose position sizes in the filing, but the shift marks a clean break from precious metals into compute infrastructure plays.
The timing is surgical. Bitcoin miners faced margin compression through 2023 and early 2024 as hashrate difficulty climbed and spot prices stalled. A subset—CoreWeave, Hut 8, and smaller names—responded by converting existing datacenter shells and stranded power contracts into GPU clusters for AI inference and training workloads. Third Point's 13F suggests the firm sees more upside in the salvage arbitrage than in gold's traditional safe-haven bid, even as macro volatility typically lifts both.
The gold exit is the louder signal. Third Point held the position through regional banking stress in Q1 2023 and the initial Fed pivot signals in late 2023. Unwinding now—while Bitcoin rallies and equities chop—suggests Loeb views the miner-to-AI thesis as a better-structured asymmetry. These companies inherit purpose-built power infrastructure, favorable utility agreements, and physical footprints in low-cost jurisdictions. The pivot to AI inference sidesteps hashrate competition and taps enterprise gross margins 30-50 bps higher than crypto mining. For a distressed-oriented fund, the setup resembles buying post-restructuring equity in energy companies with stranded LNG terminals later repurposed for hydrogen.
Third Point's portfolio construction typically runs 12-18 core positions with 3-5 opportunistic sleeves. The miner-to-AI bet likely sits in the latter bucket—small enough to contain idiosyncratic risk, large enough to move the fund if infrastructure demand scales. The broader message: Loeb is pricing in sustained AI capex spend, not a 2024 pause. That puts Third Point on the same side as hyperscalers racing to lock compute capacity before training runs for next-gen models begin in Q4 2024 and Q1 2025.
Operators should watch three follow-on events. First, whether Third Point discloses specific miner names in upcoming investor letters or conference appearances—Loeb has historically telegraphed high-conviction trades within 60-90 days of initial entry. Second, whether other Tiger cubs or distressed specialists file similar 13Fs in the next reporting window, signaling a crowding trade. Third, GPU spot lease rates on secondary markets; if inference demand tightens supply, these converted miners gain pricing power faster than hyperscaler-owned infrastructure.
The 13F also clarifies where Loeb isn't betting: on gold as a hedge or Bitcoin as a direct asset. He's long the infrastructure layer beneath AI, not the commodity itself. That distinction will matter if inference workloads plateau before the next training wave or if power costs in repurposed facilities prove stickier than expected. For now, Third Point is underwriting the thesis that stranded mining rigs in Texas and the Dakotas are the cheapest way to own AI compute exposure. The next quarterly filing will show whether that conviction held through summer volatility.