Ramp AI Index: Top Enterprises Cut Per-Employee AI Costs by 9.7% in August Amid Falling Token Prices

Median per-employee AI spending at the top 1% of US companies, which drive the bulk of enterprise AI revenue, fell 9.7% in August 2026 to $7,205, according to Ramp's AI Index for September 2026. This decline signals a strategic shift among leading AI adopters toward cost optimization, driven by falling token prices and a migration to cheaper standard AI models.

Shifting Dynamics in Enterprise AI Spending

Despite the reduction in per-employee spending among the top tier, AI adoption generally continued its upward trajectory. Ramp's data shows that 43.8% of U.S. companies on its companies on its platform utilized Anthropic services i utilized Anthropic services in August, a slight increase of 0.34 percentage points. Similarly, OpenAI's reach expanded to 39.8% of companies, up by 0.09 points. This indicates that while more companies are integrating AI, the largest spenders are becoming more discerning about their investments.

The primary drivers behind this spending reduction are twofold: a substantial drop in token prices and a noticeable migration from expensive frontier models to more cost-effective standard alternatives. The effective price per million tokens has fallen by 41% since its peak in March 2026, now standing at $0.68. This price adjustment significantly impacts the operational costs for heavy AI users.

The Rise of Standard Models and Cost Optimization

Companies are increasingly shifting workloads away from high-cost frontier models such as Opus, Fable, and Sol. The token share for these advanced models declined from 53% in early August to 45% by early September. This volume growth is largely being absorbed by cheaper standard models, including GPT-5.6 Terra and Claude's Sonnet series, which are now perceived as sufficiently capable for a broader range of enterprise applications.

Ara Kharazian, Ramp's chief economist, suggests that part of the August decline might be seasonal, attributing it to engineers taking vacations. However, he also points to a more structural change: companies are implementing internal policies that restrict the use of frontier models. This indicates a growing confidence in the performance of standard models for everyday tasks, reducing the perceived necessity for premium, higher-priced options.

Why Open-Weight Models Aren't Driving the Shift

Interestingly, open-weight models do not appear to be a significant factor in this cost-saving trend. According to Ramp's data, only 6.4% of AI-using companies on their platform run open-weight models, and across all companies, this figure drops to 3.6%. The actual adoption might be even lower, as Ramp's measurement includes usage through routing platforms that also support closed models. This suggests that the current cost optimization is primarily occurring within the ecosystem of commercial, closed-source models.

Implications for AI Providers and Enterprise Strategy

This data from the Ramp AI Index serves as a critical signal for AI providers. It suggests that the largest enterprise buyers are prioritizing cost efficiency over simply scaling their investment in the most advanced, and often most expensive, frontier models. This challenges the assumption that enterprises will consistently pay a premium for top-tier models, indicating a potential ceiling in corporate willingness to invest in the highest-cost AI solutions.

Kharazian previously highlighted

Sources

• Anthropic now has more business customers than OpenAI, according to Ramp data | TechCrunch
• AI News Center (2026) | Insights, Headlines & Podcast | Best-AI.org
• September 2026 Ramp AI Index: Cracks in the AI thesis, part 2
• Top AI spenders cut per-employee costs by nearly 10 percent in August
• AI Pricing Intelligence - Compare 1,837+ Tools | Best-AI.org