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

A recent analysis from Andreessen Horowitz (a16z) highlights a Jevons paradox unfolding in the AI market. As the cost per AI token continues to decline, overall demand for compute is growing faster than prices fall, resulting in H100 GPU rental prices holding steady or even increasing. This dynamic means that cheaper inference is not translating into lower infrastructure costs for cloud providers.

The phenomenon is significant because it suggests that the AI market is still in a phase of explosive demand expansion. Organizations are consuming more tokens than ever, driven by broader adoption of LLMs across enterprise and consumer applications. The falling unit cost of AI is acting as a demand accelerator rather than a cost deflator.

For NVIDIA and Jensen Huang, this scenario represents the best-case outcome: sustained and growing demand for AI accelerators despite falling token prices. If demand were to flatten, the entire chain from chip manufacturers to cloud providers would face pressure, as revenue would depend on volume growth alone.

The key risk identified is that if demand growth slows, the economics of the AI infrastructure stack could deteriorate. Cloud providers would face a squeeze between rising GPU costs and competitive pressure to lower token prices. The current trajectory, however, suggests the market is still in a phase where demand outpaces supply-side cost reductions.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.