Peter Thiel's investment vehicle filed a 8-stock 13F for Q4 2024, and seven of the eight positions are regulated utilities. No semiconductor names. No cloud infrastructure. No AI darlings. The portfolio composition is deliberate: the next constraint in artificial intelligence is not model architecture or chip supply—it is gigawatt-scale power delivery to data center clusters already under construction.
The filing shows concentrated exposure to names like NextEra Energy, Duke Energy, and Southern Company—operators with multi-gigawatt baseload capacity and interconnection pipelines stretching into 2027. Thiel's vehicle held no positions in Nvidia, AMD, or any hyperscaler. The lone non-utility position was a small stake in a defense contractor with grid-hardening contracts. This is not a diversified technology bet. It is a single thesis: AI inference at scale requires stable, contracted electricity in 100+ megawatt increments, and utilities with generation assets and regulatory capture will extract the rent.
The timing aligns with three separate developments in the past six months. First, Microsoft announced 20-year power purchase agreements with two nuclear operators in Pennsylvania and Virginia, signaling that hyperscalers are locking in offtake before data centers break ground. Second, Dominion Energy disclosed that 60 percent of its new interconnection requests in Q3 came from AI-adjacent industrial users, not residential or commercial growth. Third, PJM Interconnection—the largest grid operator in North America—revised its interconnection queue processing time upward to 48 months for requests filed after January 2024, creating a structural moat for utilities with existing capacity.
The portfolio structure suggests Thiel is not betting on a single utility outperforming. He is betting that the entire sector will reprice as AI operators realize power is the gating constraint. Data center construction timelines are now 18 to 24 months. Substation upgrades and new transmission lines take 36 to 60 months. The arb is obvious: equity markets are pricing utilities as low-growth regulated monopolies yielding 3 to 4 percent, while AI operators are negotiating offtake agreements at premiums to wholesale rates and committing to 15 to 20-year terms. If even 10 percent of planned AI data center capacity comes online by 2027, utilities with flexible generation assets will see contracted revenue growth that has not been modeled into current valuations.
Allocators should track three signals in the next twelve months. First, watch PJM and ERCOT's quarterly interconnection reports—if AI-related requests exceed 15 percent of total queue volume, it confirms Thiel's thesis is consensus. Second, monitor earnings calls from NextEra, Duke, and Southern for mentions of long-term industrial offtake agreements; any disclosed terms above $70 per megawatt-hour suggest pricing power is shifting. Third, watch for secondary offerings or debt issuances from utilities—if they are raising capital outside normal maintenance cycles, it signals accelerated capex to meet AI demand. These are not 2025 events. They are 2026 to 2027 markers.
Thiel filed this 13F in February 2025, and the positions were held as of December 31, 2024. If the thesis plays out, the next filing will show size increases, not rotation.