June 10th news, a new research report released by Ramp AI Index shows that the adoption rate of artificial intelligence by U.S. companies is in a continuous growth phase. Among the top 1% "AI experts (AI advanced users)", companies have invested up to $7,500 per employee per month in AI. Although this figure is still lower than the average monthly salary of software engineers at about $16,000, the rapid growth in enterprise AI spending has sparked widespread discussion in the industry about the structure of computing power and labor costs.

Artificial Intelligence AI Robot (2)

Previously, NVIDIA executives had publicly stated that computing costs have now exceeded employee salaries. The CEO of AI startup Mercor also revealed that their internal token (Token) expenditures have exceeded the total number of employees, indicating that in some leading companies, the cost of computing power is catching up with or even surpassing labor costs.

Looking at the overall market, enterprise AI spending shows a polarized characteristic and strong growth momentum. Among companies that have adopted AI technology, the average AI spending per person increased by 14.1% compared to the previous month. However, compared to the massive investment of the top 1% companies at $7,500 per month, the spending per employee for the top 10% of users dropped sharply to about $611 per month, while the median user spends around $11.38 per person per month, which is only equivalent to the cost of one seat in an enterprise software package. This means there is still great potential for deep penetration of AI in enterprises.

Faced with the pressure of continuously consuming token budgets, the top 1% companies currently tend to adopt flexible "hybrid" strategies, switching flexibly between multiple cutting-edge models and platforms, and introducing more economical open-source models to balance costs. This action of seeking the optimal cost-effectiveness among multiple models not only reflects the rational choices of companies in dealing with rising computing power costs, but will also further accelerate the evolution of the AI infrastructure layer towards multi-modal, refined, and high-cost-effectiveness directions.