The AI Economy: Why Less Expensive Tokens Still Mean Higher Bills

Investing.com — The recent drop in prices for AI services has not dampened spending on AI computing power; instead, it is driving companies and individuals to use more, creating a feedback loop that could keep technology investment high.

According to UBS, the price of AI tokens has fallen by approximately half since May, as the LLM Token Expenditure Index declined from nearly $2 per million tokens to about $1 by August. Demand has meanwhile quickened, illustrating the Jevons paradox—where efficiency gains reduce costs sufficiently to fuel greater consumption.

UBS notes the trend is especially evident on OpenRouter, a gateway to hundreds of large language models. Token usage on the platform has grown tenfold since January, while average monthly outlays among its most active 1% of users climbed to $7,500 from $2,500.

The widening adoption of AI is also underpinning the wider investment cycle. As businesses increasingly gauge AI workloads in tokens, the uptick in token usage bolsters the rationale for accelerating capital expenditure on data centres, computing hardware and related infrastructure.

Token demand can grow through extended prompts and outputs, multimodal applications, search and retrieval, and more involved reasoning, UBS said. AI agents capable of employing tools, devising plans and carrying out tasks without human supervision can lift token usage even further.

Efficiency improvements are reinforcing that trend. Hardware and AI models have grown more efficient; an Economist study referenced by UBS found that the token cost of reaching chosen performance benchmarks dropped roughly 97% year over year. Cheaper pricing and promotional incentives from major AI model developers may quicken adoption even more.

This dynamic is turning what UBS calls an “AI flywheel”: lower-cost AI fosters broader usage; higher usage drives demand for computing capacity; and that demand encourages additional investment. A chart on page 3 depicts token prices declining sharply over the summer even as memory prices rose, highlighting the growing gap between more affordable AI use and climbing infrastructure expenses.

For investors, token pricing and consumption have become key gauges of AI-economy activity, according to UBS. Instead of pointing to weaker spending, declining token prices may indicate that AI adoption is expanding swiftly enough to lift total token usage and expenditure.