Cheaper AI tokens are driving more demand, and that's Jensen Huang's best-case scenario

Your cloud GPU costs are unlikely to fall as quickly as AI model inference prices do.

A phenomenon dubbed the "Jevons paradox" is observed in the AI market, where decreasing token prices are paradoxically leading to increased demand and sustained or rising GPU rental prices, according to data through August 2026 cited by a16z. Cheaper AI tokens are enabling a surge in AI agents, automation, and novel applications, causing usage volume to outpace the rate at which per-unit costs are falling. The exact split between human-driven and system-driven AI demand remains unclear, though agentic AI is known to consume tokens at an exceptionally high rate. This dynamic suggests that compute demand might be artificially inflated, with even minor increases in human usage potentially driving significant hardware requirements. The entire AI ecosystem's stability hinges on AI usage growing sufficiently to counteract falling token prices, which in turn supports high hardware costs and scarcity. Any flattening in demand could negatively impact the entire supply chain, from chip manufacturers to cloud providers, with ripple effects already visible in market reactions to revenue reports.

The Jevons paradox, where increased efficiency leads to increased consumption, is fundamentally reshaping the AI hardware market. While falling token prices make AI more accessible and drive unprecedented usage growth, this very demand is sustaining or even increasing the rental costs for high-end GPUs like the H100. This indicates that the bottleneck is not just the cost of individual AI computations, but the sheer volume of compute required, which is rapidly expanding with agentic AI and new applications. The industry's economic model is predicated on this escalating demand offsetting falling per-token costs, creating a delicate balance that supports current hardware valuations and scarcity.

This trend has significant implications for companies across the AI value chain. Chip manufacturers and cloud providers benefit from sustained high demand for compute, but the AI ecosystem's vulnerability to any slowdown in usage growth is starkly evident. If demand falters, the chain reaction could impact energy providers, memory suppliers, and even broader financial markets, as demonstrated by market sensitivities to AI revenue figures. Consequently, future investments and strategic planning must account for this paradoxical loop where lower AI operational costs directly fuel a more intense demand for underlying hardware infrastructure.