Thesis
The Post-Software Cycle
The marginal value of AI is transitioning from software models to physical-world deployment. The bottlenecks are now energy and infrastructure, not code.
Software-only moats compress
Every capable lab can now train a frontier model. Open weights close the gap. The differentiation runs out.
Physical-input moats compound
You cannot open-source a power plant. You cannot clone a rare-earth refinery. Capital that owns the physical substrate of AI deployment captures outsized returns as the industry matures.
The Age of Abundance
When AI bridges into humanoids, automated factories, and autonomous systems, the first country to achieve this lowers labor costs tenfold. Physical AI is the next moat, not model quality.
What we back
- Electric utilities with secured PPAs
- Rare earth producers and allied refiners
- Japanese + European industrial automation
- SMR developers and uranium fuel chain
- Brazil exposure — Compute Antarctica
- Orchestration platforms for physical AI