AI inference costs plunge 47% per quarter

Hyper-deflation in AI inference costs threatens to upend current capital expenditure assumptions—if raw intelligence rapidly approaches zero marginal cost, massive data center moats may provide diminishing returns far faster than investors anticipate.

• –**Unprecedented Deflationary Velocity:** A 13-fold annual reduction in inference cost dwarfs Moore's Law, fundamentally shifting the bottleneck from hardware availability to application-layer integration.
• –**Rapid Reasoning Commoditization:** Benchmarks such as GPQA Diamond demonstrate that frontier-grade reasoning becomes micro-penny commodity compute in under two years.
• –**Evolving Open vs. Closed Dynamics:** Proprietary labs are competing aggressively on inference efficiency rather than raw capability alone, narrowing the operational cost window that open models traditionally exploited.
• –**Edge Viability Approaching:** As compute efficiency accelerates, models matching current frontier capabilities will feasibly run on standard consumer hardware within three to five years, dampening long-term cloud dependency.