Data indicates that lower AI token costs are fueling higher demand for H100 GPUs, a trend described as Jevons' paradox.
Key facts
- •Data from Ornn, Silicon Data, and Bloomberg shows a Jevons paradox in the AI market as of August 2026.
- •H100 GPU rental prices have remained steady or climbed despite falling token prices.
- •Agentic AI is driving high token consumption, potentially inflating compute demand.
- •The hardware market relies on AI usage growth to maintain high demand for chips and energy.
- •U.S. stock markets showed sensitivity to reports concerning OpenAI's annualized revenue.
Data from Ornn, Silicon Data, and Bloomberg as of August 2026 suggests the AI market is experiencing Jevons' paradox. While the price of AI tokens continues to decline, the rental costs for H100 GPUs remain steady or are increasing. This dynamic suggests that cheaper tokens are unlocking new applications and automation, causing volume growth to outpace the reduction in per-unit costs.
The impact of agentic AI
The rise of agentic AI, which consumes tokens at a high rate, is a significant factor in current demand. It remains unclear how much of this consumption is driven by human users versus automated systems. This high volume of usage is currently keeping hardware scarce and expensive.
Market risks and dependencies
The stability of the current AI ecosystem depends on usage growth continuing to offset falling token prices. If demand for AI services flattens, the entire supply chain—from chip and memory manufacturers to energy and cloud providers—could face negative impacts. Market sensitivity to these trends was recently demonstrated when U.S. stocks declined following reports regarding OpenAI's annualized revenue.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by The Decoder.
