U.S. companies continue to increase their investments in AI, but overall data shows that AI costs have not yet surpassed labor costs. According to the latest Ramp AI Index, the top 1% of most aggressive adopters are spending an average of $7,500 per employee per month on AI.
The top 1% of companies spend more.
This level is significantly higher than that of ordinary companies but still below the average monthly salary of software engineers, which is about $16,000. As companies continue to consume model invocation and computing power budgets, the market is beginning to ask whether firms are already spending more on AI than on their employees.
Ramp refers to these highly AI-intensive companies as "AI-pilled" businesses. Data shows that per-employee AI spending for these companies increased by 14.1% month-over-month last month, indicating that investment in this area continues to expand.
- Top 1% of companies: Average monthly per capita of $7,500
- Top 10% of companies: Average monthly per capita of $611
- Median enterprise: average monthly per capita of $11.38
Top users increase overall investment.
The report notes that recent corporate executives have publicly highlighted rising AI costs. An executive at NVIDIA stated that compute costs have surpassed employee salaries. The CEO of recruiting startup Mercor also said that the token costs consumed by the company’s internal AI agents now exceed human labor expenses.
However, looking at a broader sample of companies, this has not yet become a widespread phenomenon. The current rapid growth in AI budgets is still primarily concentrated among the heaviest users, rather than across the entire corporate population.
Enterprises shift to a multi-model portfolio
Ramp's research also shows that enterprises with the highest AI usage typically do not rely solely on a single model or platform, but instead switch between multiple cutting-edge models while integrating lower-cost open-source models to reduce usage costs.
This means that while companies are scaling up their use of AI, they are also adjusting their procurement methods. Whether such spending continues to grow rapidly in the future will depend on model prices, computing costs, and the speed of internal application deployment.