Cheaper AI tokens are fueling higher demand, with token prices continuing to decline while H100 GPU rental prices remain steady or rise, according to data from Ornn, Silicon Data, and Bloomberg as of August 2026. This trend aligns with what a16z describes as a textbook example of Jevons' paradox in the AI market.
The market dynamics suggest that as token costs drop, the volume of AI applications and usage grows at a faster rate than the decline in per-unit costs. This growth is attributed to the emergence of AI agents and automation, which are unlocking new use cases and increasing overall demand.
However, the distinction between demand driven by human users and that generated by agentic AI systems remains unclear. AI agents are consuming tokens at an accelerated rate, which could artificially inflate compute demand and lead to significant hardware requirements even with modest human usage growth.
"Jevons' paradox is Jensen's best friend," said Matthias Bastian. The statement highlights the belief that as token prices fall, AI usage must grow fast enough to sustain hardware scarcity and high prices. If demand fails to grow, it could trigger a cascade of economic impacts across the supply chain.
The current scenario is predicated on the assumption that AI usage will continue to expand rapidly. If this growth stalls, the ripple effects could be severe, impacting everything from chip manufacturers to energy providers and cloud companies. Recent market volatility, such as the drop in US stocks due to concerns about OpenAI's revenue, underscores the sensitivity of these markets.
Jensen Huang did not specify how much of the demand comes from human users versus AI systems, and the source raises the question of whether compute demand is being artificially inflated. The next step, according to the source, is to monitor how the AI market evolves in response to these trends.
Source: thedecoder
