2026-09-11 · Fuente：Unite AI
Per-million-token pricing across leading models has collapsed more than 90% in two years, but a new analysis argues unit-price declines are not translating into lower enterprise AI spending.
The economics conversation surrounding enterprise generative AI has converged on a single number: the price of LLM inference. Across the industry's leading models, the cost per million tokens has fallen by more than 90% over the past two years — a decline so steep that many corporate leaders have concluded the financial side of generative AI is safely under control. A new analysis from Unite AI argues that this assumption deserves scrutiny, because falling unit prices are not translating into lower enterprise AI bills.
The price collapse itself is not in dispute. The report characterizes the reduction in per-token pricing as a genuine milestone for the industry, one that allows companies to run intelligence at a fraction of what it cost a year earlier. On paper, the trajectory looks like an unambiguous win for adopters: the same budget buys dramatically more model output than it did two years ago.
The gap, according to the analysis, opens between the sticker price and the invoice. The report points to how tokens are actually consumed in production — including token waste and the costs of governing context in agentic AI workflows, the themes at the core of the piece — as the reason total bills resist decline even as per-token rates plunge. In other words, the metric dominating boardroom discussion, price per million tokens, captures only one slice of the cost structure enterprises actually face.
For AI-compute buyers and investors, the implication is that inference-pricing headlines are an increasingly unreliable proxy for enterprise AI economics. A decline of more than 90% in token prices over two years is real and consequential, but the report's central point is that it has not delivered proportionate relief on corporate bills. The number that matters for enterprise finance teams, the analysis suggests, is the gap between unit cost and total spend — not the headline price per million tokens.
Etiquetas: LLM Inference · Token Pricing · Enterprise AI · Agentic AI · Context Governance · AI Economics
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado