According to corporate card and spend platform Ramp’s September 2026 AI Index, median per-employee AI spending among the top 1 percent of enterprise spenders fell 9.7 percent in August to $7,205. While overall business adoption of AI tools continued to grow slowly, top-tier enterprise clients cut expenditures due to falling API token prices and strict internal policies steering usage toward standard models.
The report highlights that effective price per million tokens dropped 41 percent from its March 2026 peak down to $0.68. As a result, volume consumption for frontier models like Opus, Fable, and Sol fell from 53 percent at the start of August to 45 percent in early September, while cheaper alternatives like GPT-5.6 Terra and Claude Sonnet captured market share.
Notably, the transition away from frontier closed models is not primarily benefiting open-weight options, as only 6.4 percent of AI-using companies on Ramp run open-weight models. The overall spend reduction highlights emerging headwind risks for frontier model providers relying on premium API pricing models to sustain revenue growth.
Why it matters
Enterprise buyers are enacting strict cost controls, prioritizing cheaper ‘good enough’ standard models over high-margin frontier reasoning APIs.
Frontier model providers face margin pressure as declining token prices outpace volume growth ahead of planned public listings like Anthropic’s.
Open-weight adoption remains under 7% among spending enterprises, leaving commercial API routing platforms in firm control of business workloads.
Source: the-decoder.com
