The Decoder reported on 11 October that AI token prices keep falling, yet renting H100 GPUs costs the same or more. A token is a small piece of text that an AI model reads or writes. The outlet cites data from Ornn, Silicon Data and Bloomberg, as of August 2026. It says investment firm a16z calls this a textbook Jevons paradox. Cheaper tokens make AI agents, automation and new apps worthwhile, so total use grows faster than the price per token drops.
The Decoder says it is not clear how much of the demand comes from people and how much from the AI systems themselves. It notes that AI agents use tokens at a very high rate. It says compute demand could be inflated, and that modest growth in human use might still call for far more hardware. The outlet adds that the pattern holds only if usage keeps growing fast enough to cover falling prices. It gives no usage figures or price levels. It also does not say how the OpenAI revenue reports that moved US stocks relate to these numbers.
For teams that buy AI services or run their own models, the data suggests that cheaper tokens do not mean cheaper GPU capacity. The Decoder's figures show H100 rental prices holding steady or rising while token prices fall. Budgets for AI workloads may need to track both numbers.