Dan Loeb's Third Point has taken stakes in Bitcoin mining operators pivoting their data center assets toward AI compute and inference workloads. The firm disclosed positions in at least three companies—publicly traded names repositioning stranded GPU capacity originally deployed for proof-of-work mining—representing an estimated $50 million to $80 million in aggregate exposure. The investments span infrastructure operators with existing co-location agreements and power contracts in Texas, North Dakota, and Wyoming, where electricity costs remain below 4.2 cents per kilowatt-hour.
The pivot mirrors a structural arbitrage Third Point identified six months ago: mining operators hold long-term power purchase agreements locked at pre-2022 rates, purpose-built cooling systems designed for high-density compute, and real estate already zoned for industrial power loads. As Bitcoin mining economics compressed—network hashrate climbed 42% year-over-year while block rewards halved in April 2024—operators faced a binary choice between liquidation or asset redeployment. Third Point's thesis hinges on the spread between mining profitability, which fell to $0.08 per terahash daily, and AI training economics, where Nvidia H100 clusters generate $2.30 to $3.10 per GPU-hour depending on utilization and customer tier.
Loeb's positioning reflects the fund's broader belief that frontier model training will migrate toward purpose-built edge compute facilities rather than hyperscaler data centers. The bottleneck is not chip supply—Nvidia shipped 550,000 H100-equivalent GPUs in Q4 2024—but rather power allocation and cooling density. Former mining sites offer pre-approved electrical substations capable of scaling to 50 to 150 megawatts without multi-year permitting delays. One Third Point portfolio company reportedly secured a $120 million equipment financing facility in December to retrofit two mining facilities with liquid-cooled GPU racks targeting mid-2025 deployment. The company's forward order book includes contracts with two AI research labs and one defense-adjacent entity, all under NDA.
Allocators should monitor three specific developments. First, whether these operators can secure Nvidia H100 or H200 allocations at enterprise pricing—current lead times remain 16 to 22 weeks for non-hyperscaler buyers. Second, track whether power utilities in mining-heavy states renegotiate industrial rate schedules; Texas regulators are reviewing load-balancing agreements that currently exempt data centers from peak-hour curtailment. Third, watch for debt restructurings among publicly traded miners who did not pivot—at least four companies face maturity walls in Q2 and Q3 2025, creating potential distressed acquisition opportunities for infrastructure specialists.
Third Point's stake sizes remain undisclosed pending 13F filings due mid-February, but the positioning occurred during Q4 2024 alongside exits from two legacy SaaS holdings. The fund's technology book now tilts toward infrastructure plays with embedded real asset optionality—compute, power, and land—rather than pure software multiples.