The AI Economy: Why cheaper tokens still lead to bigger bills

The cost of artificial intelligence (AI) services has dropped significantly, yet companies and users are consuming more AI computing power rather than spending less. This trend, known as the Jevons paradox, shows that increased efficiency can lead to greater overall consumption. On the platform OpenRouter, token usage has surged tenfold since the start of the year, with the top 1% of users now spending an average of $7,500 per month, up from $2,500.

This growing demand for AI is driving further investment in the technology. Companies are now measuring AI workloads in tokens, and as token consumption increases, it strengthens the case for spending on data centers, computing equipment, and other infrastructure. AI usage can rise through longer prompts and responses, multimodal applications, search and retrieval, and more intensive reasoning.

AI agents that can perform tasks without human oversight are also contributing to higher token consumption. Efficiency gains in hardware and AI models have made them more efficient, while AI model providers offering incentives could further accelerate adoption. This creates a positive feedback loop, or "AI flywheel," where cheaper AI enables more usage, higher usage supports demand for computing capacity, and stronger demand encourages further investment.