2 months ago
- Compute is becoming an indispensable resource akin to 'spice' in Dune, powering AI and constraining growth due to supply chain bottlenecks.
- AI workloads, especially agentic systems, now consume 100x to 1,000x more tokens than traditional chatbots, accelerating demand.
- Multiple simultaneous bottlenecks exist: GPU scarcity, memory price surges, TSMC capacity sold out, and power infrastructure constraints.
- Scarcity is repricing assets, such as bitcoin miners with power and grid access, which are now seen as strategically valuable 'Lazarus assets'.
- Access to compute is becoming a competitive moat, with companies securing long-term supply gaining advantages over those without.
- Higher inference costs may slow AI adoption, concentrating resources on high-value applications and disadvantaging smaller firms.
- The AI race has shifted from a software competition to an industrial-scale infrastructure race for chips, memory, electricity, and land.