Cheaper AI has not meant smaller AI bills. Data published by venture firm a16z in its Charts of the Week on 31 October 2026 shows average AI token prices fell 67% since May 2026, while total token spending climbed between 122% and 160% over the same stretch.
What is an AI token?
A token is the billing unit for AI model inference — a small chunk of text a model reads or writes — and providers charge users by the token. So a two-thirds price fall paired with spending more than doubling means usage grew fast enough to swamp the price cut entirely.
The agent driver
The biggest driver appears to be autonomous AI agents. According to the a16z data, agents consume nearly five times as many tokens as human users. A person asks a chatbot a question and reads the answer; an agent runs a longer task, carrying extensive context and taking many steps, and every step burns tokens. Agent token usage has also grown roughly 14 times since February 2026, pointing to a shift toward larger, automated workflows rather than one-off questions.
The Jevons paradox
A16z attributes the pattern to the Jevons paradox — when a resource becomes cheaper and more efficient to use, total consumption rises rather than falls, because lower cost opens up uses nobody bothered with before.
Hardware costs are rising too
Rental costs for Nvidia’s B300 chips rose 106% since mid-July 2026, according to the data, while B200 rentals climbed 66% over the same period — a reminder that cheaper model calls do not mean cheaper infrastructure. A relatively small group of top users accounts for a significant share of total token spending. Related AI-spending context is in Emadan News’ reporting on Anthropic’s pre-IPO derivatives valuing the firm at $1.9 trillion, which shows how much appetite there is for exposure to leading AI companies.
What it does not prove
Spending growth is not profit, and it is not investment advice. The summaries of the a16z data do not publish the firm’s full methodology or a precise definition of the ‘AI token spending’ metric, so the 122% to 160% range should be read as a reported range from the underlying data, not a precise audited total. The durable takeaway is the direction — usage up sharply while unit prices fell — not any single decimal point.
Frequently asked questions
What is an AI token? It is the billing unit for AI model inference: a small chunk of text a model reads or writes, and providers charge by the token.
Why would spending rise if prices fall? Because usage rose even faster than prices fell. Cheaper tokens, especially through autonomous agents running long, multi-step tasks, opened up far more use — the Jevons paradox described by a16z.
Does higher token spending mean AI companies are profitable? No. Spending is a measure of usage and bills, not profit, and the reported range is not an audited total.
Sources
- CryptosBank: AI token spending jumps 160% as inference prices fall
- CryptoBriefing: a16z AI token spending surges as prices fall
- SpendNode: AI token spending surges, a16z November 2026
