Top AI spenders cut per-employee costs by nearly 10 percent in August
decoder ·
Ramp's AI Index for September 2026 shows that US business AI adoption is still growing, but more slowly. Among the top 1 percent of AI spenders, median per-employee spending fell 9.7 percent in August to $7,205, while spending at the top 10 percent and the median continued to rise. Anthropic was used by 43.8 percent of US companies, up 0.34 percentage points, and OpenAI reached 39.8 percent, up 0.09 points, with IT and finance still leading adoption. Ramp chief economist Ara Kharazian attributed part of the decline to August vacations, but also pointed to falling token prices and a shift toward cheaper models. The effective price per million tokens dropped 41 percent from its March 2026 peak to $0.68, and OpenAI has consistently been cheaper than Anthropic. Usage growth is mainly coming from cheaper standard models such as GPT-5.6 Terra and Claude's Sonnet series, while frontier models like Opus, Fable, and Sol saw their token share fall from 53 percent in early August to 45 percent in early September. Companies are adopting internal policies that limit use of expensive frontier models because standard models are seen as good enough and much cheaper. Open-weight models still have little impact: only 6.4 percent of AI-using companies on Ramp run them, and just 3.6 percent of all companies do. Ramp's data covers only part of the market, and the picture may become clearer when Anthropic reportedly files for an IPO planned for October.
decoder