Ramp's September AI Index shows top-spending firms cut per-employee AI costs ~10% in August as token prices fall and usage shifts to cheaper models

Ramp's September AI Index, built from its US business card and spend data, reports that median per-employee AI spend among its top 1% of spending companies fell 9.7% in August to $7,205. It also found the effective price per million tokens down 41% from a March 2026 peak to $0.68, and frontier models' share of tokens consumed dropping from 53% in early August to 45% by early September. Anthropic's business adoption edged up to 43.8% of US companies and OpenAI's to 39.8%, both marginal gains.

If sustained, a shift away from frontier models toward cheaper "good enough" options would squeeze margins for providers like OpenAI and Anthropic just as Anthropic is reportedly heading toward an October IPO, and it would matter to enterprises weighing AI budget allocation. But this is one month of data from a single, SMB-skewed segment (Ramp's card customers), excluding larger enterprise and cloud-committed spend, so it is not yet evidence of a broader trend.

Ramp's own chief economist flags the top-1% segment as volatile and prone to revision, and August has a plausible seasonal effect, yet the framing ("cracks in the AI thesis") presents this as a clear inflection. Median and top-10% spend reportedly kept rising in the same period, which undercuts the headline narrative, and the claim that volume growth won't offset price declines is asserted, not shown.

1. 01Ramp's October Index for whether the top-1% spend drop persists or reverses (the report itself flags August as seasonally noisy and revision-prone)
2. 02Median and top-10% spend trends, not just the top-1% segment, since the wider cohort reportedly kept rising even in August
3. 03Anthropic's reported October IPO filing and any disclosed revenue or usage figures that either confirm or contradict a frontier-spend slowdown