The biggest buyers of artificial intelligence are pulling back. According to Ramp’s September 2026 AI Index, the top 1 percent of US companies by AI spend cut their median per-employee costs by 9.7 percent in August, dropping to $7,205. This isn’t just a seasonal dip caused by engineers taking summer vacation, though Ramp chief economist Ara Kharazian admits that plays a part. The real story is a structural shift in how enterprises view value. Companies are no longer blindly chasing the most powerful models available. Instead, they are actively restricting access to expensive frontier options like Opus and Fable in favor of cheaper, standard alternatives.
Price pressure is accelerating this trade-down. The effective cost per million tokens has plummeted 41 percent since its peak in March 2026, now sitting at just $0.68. Both OpenAI and Anthropic have joined the race to the bottom with recent price cuts, but volume growth isn’t keeping pace. Usage is migrating toward models like GPT-5.6 Terra and Claude’s Sonnet series, which enterprises deem “good enough” for most tasks. Consequently, the share of tokens consumed by frontier models fell from 53 percent in early August to 45 percent in early September. This suggests that for many corporate workflows, marginal gains in reasoning capability no longer justify the premium price tag.
Interestingly, this cost-cutting spree isn’t driving adoption of open-weight or Chinese alternatives. Only 6.4 percent of AI-using companies on Ramp’s platform utilize open-weight models, a figure that likely overstates actual market penetration since it excludes direct API usage. The shift is happening entirely within the walled gardens of major providers. With Anthropic reportedly planning an IPO for October, these metrics offer a crucial reality check for investors. The narrative of infinite enterprise demand for high-end AI is cracking. If token prices continue to fall while usage stabilizes on mid-tier models, revenue growth for model providers may face significant headwinds despite broader adoption.
