Dan Loeb's Third Point disclosed new positions in former cryptocurrency mining operations now redirecting GPU capacity toward AI inference and training workloads. The 13F filing, covering the quarter ended December 31, marks the first institutional signal that billionaire capital views salvaged mining infrastructure as viable compute arbitrage against hyperscaler bottlenecks. Third Point's entry follows 18 months of miner pivots, during which companies including Core Scientific and Hut 8 shifted from proof-of-work validation to contracts with AI labs and model deployers. The filing does not specify dollar allocation, but Third Point's recent equity book averaged $4.2B in long positions.
The thesis rests on two structural shifts. First, Bitcoin's April 2024 halving cut block rewards from 6.25 BTC to 3.125 BTC, rendering older farms uneconomical at prevailing hash rates. Second, AI compute demand outpaced supply by an estimated 40% in 2024, per Bain infrastructure analysis, creating spot pricing power for anyone holding Nvidia H100 or A100 clusters. Former miners own data centers already zoned for industrial power loads, with redundant cooling and fiber backbone in place. Retrofitting a 50 MW mining hall for AI costs roughly $18M versus $120M for greenfield builds, according to JLL Data Center Economics. That gap explains why operators like Applied Digital and Iris Energy now report GPU hosting margins near 22%, compared to 8-11% in traditional colocation.
The move matters because it confirms hedge fund interest in compute infrastructure as a tradable asset class, separate from semiconductor exposure or cloud service multiples. Third Point's prior activism in tech included Yahoo, Softbank, and Intel—positions built on misallocated capital and hidden operational leverage. Here, Loeb appears to be betting that ex-miners will either secure long-term contracts with OpenAI-tier customers or become acquisition targets for hyperscalers seeking fast capacity. Microsoft already leases 200 MW from bankrupt Compute North assets. Amazon's recent $650M outlay for a Susquehanna nuclear-powered data center signals willingness to buy rather than build when time-to-deployment matters. If Third Point's new holdings include names like TeraWulf or Cipher Mining, both of which pivoted in late 2024, the implicit valuation suggests the market is pricing these operators at 0.4x replacement cost of their GPU-ready square footage.
Operators and allocators should monitor three developments over the next 90-120 days. First, Q1 earnings from the pivoted miners will reveal whether AI hosting contracts carry take-or-pay clauses or remain spot-rate exposed. Second, Nvidia's April data center event may clarify whether B200 deployments tighten further or if supply normalization begins by mid-year. Third, any follow-on 13F filings from peer funds—Tiger Global, Coatue, or Millennium—will indicate whether Loeb's position reflects consensus or contrarian timing. If two or more funds trail into the same names, the trade becomes crowded, and early premium compresses.
Third Point entered when the pivot narrative was still unproven, which means either Loeb has term sheets the market does not, or he is pricing in a 24-month exit window where hyperscaler M&A becomes the liquidity event. The filing itself is the forward guidance.