Dan Loeb's Third Point disclosed equity positions in former cryptocurrency mining operations that have redirected data center capacity toward artificial intelligence workloads, according to recent 13F filings. The hedge fund's investments target companies that inherited tens of thousands of graphics processing units and power infrastructure originally built for proof-of-work blockchain validation.

The pivot follows a 60-70% collapse in Bitcoin mining profitability since the April 2024 halving event, which cut block rewards from 6.25 to 3.125 BTC. Mining operators with stranded assets in low-cost power jurisdictions—West Texas, upstate New York, Iceland—found that the same cooling systems, redundant power feeds, and high-density rack configurations required for SHA-256 hashing align closely with requirements for large language model inference and training. Third Point's entry suggests institutional validation of this asset class redeployment thesis, particularly as hyperscalers face 18-24 month lead times for purpose-built AI data centers.

The economic logic is tighter than it appears. A former mining facility in Midland, Texas, running 10 megawatts of continuous load, can convert to AI inference at $0.03-0.04 per kilowatt-hour—half the cost of traditional colocation. These sites already cleared environmental review, hold utility interconnection agreements, and maintain 99.9% uptime SLAs. Third Point likely evaluated the arbitrage between distressed mining equity valuations—many trading below net asset value—and the $250-400 per kilowatt premium that AI-focused REITs command in private markets. The fund's involvement also signals confidence that demand for edge inference, particularly in latency-sensitive applications like autonomous vehicle compute, will absorb second-tier GPU clusters that lack the NVLink density required for frontier model training.

Allocators should track three developments over the next six months. First, whether Third Point's portfolio companies secure offtake agreements with model developers outside the Magnificent Seven, particularly Asian AI labs and defense contractors building sovereign compute. Second, whether power utilities in Texas and the Pacific Northwest impose demand charges or curtailment clauses that erode the cost advantage these sites currently enjoy. Third, the pace at which Nvidia's H100 and H200 chips filter into the secondary market as hyperscalers upgrade to Blackwell architecture—flooding supply could compress inference pricing and margin assumptions embedded in these conversions.

Third Point entered when Bitcoin mining stocks were trading at 0.4-0.6x book value, and before AI infrastructure became consensus.