800 million weeks of activity, 320-fold increase in token consumption... OpenAI's latest report: Enterprise AI has entered the “deep water zone”

The Zhitong Finance App learned that OpenAI recently released the “Report on the Status of Artificial Intelligence (AI) in Enterprises”. The report shows that AI is fully penetrating all walks of life, with technology, healthcare, and manufacturing becoming the fastest-growing fields.

The tech giant revealed that this analysis data comes from two major sources: one is actual usage data from OpenAI enterprise customers, and the other is an AI application model survey conducted on 9,000 employees of nearly 100 companies.

The report also revealed that the number of weekly active users of ChatGPT has exceeded 800 million.

OpenAI emphasized that the current application of AI by enterprises is showing the dual characteristics of accelerated penetration and deep integration. Over the past year, the average weekly interaction volume of ChatGPT Enterprise Edition has increased by about 8 times, and the average weekly number of instructions sent by a single user has increased by 30%. Judging from the application model, the frequency of use of structured workflows such as project management and customized GPT has surged 19 times since this year, indicating that AI application scenarios have moved from scattered impromptu queries to systematic and reusable business processes.

The consumption of model inference computing power at the enterprise level has also risen sharply. In the past 12 months, the average inference token consumption of a single enterprise has increased by about 320 times. OpenAI explained that this data confirms that a more intelligent model is being systematically integrated into the company's continuously expanding product and service systems.

The report further indicates that AI is being used to complete increasingly complex professional tasks. Judging from the industry distribution, the popularity of AI covers all fields. Among them, the professional services, finance and technology industries have shown a particularly strong growth momentum with the advantage of large-scale implementation.

Looking at the global market, the number of enterprise-level customers in Australia, Brazil, the Netherlands and France all increased by more than 140% year over year, making it the fastest growing regional market. From a technical interface perspective, OpenAI's international application programming interface (API) enterprise customer base has increased by more than 70% in the past six months. Among them, Japan is the country with the highest number of enterprise-level API customers outside the US.

According to research data, 75% of workers in the surveyed companies said that after using AI tools, the efficiency or quality of work output improved significantly. In terms of time costs, the average user saves 40 to 60 minutes of working time per day, and high-frequency users save more than 10 hours of work per week.

It is worth noting that the number of programming-related instructions sent by non-technical employees increased by 36%; at the same time, 75% of users used AI to successfully unlock tasks that could not be completed independently in the past.

The report also revealed a key trend: the gap between frontier users, leading companies, and industry average continues to widen. From a user perspective, the average weekly number of instructions sent by leading users (top 5%) is 6 times that of ordinary users, and the advanced features of the model are used more deeply; from an enterprise perspective, the average weekly interaction volume of leading companies per seat is 2 times the industry median, and their AI technology deployment has achieved deep collaboration across teams. OpenAI points out that this gap has important practical significance — research shows that the more intelligent computing power users use and the more diverse business scenarios they cover, the more significant the savings in working hours.

In a blog post published this Monday, OpenAI emphasized: “On average, we launch a new feature or capability every three days. Today, the core bottleneck that limits enterprises to embrace AI is no longer model performance or the tool itself, but the suitability of the enterprise's organizational structure and implementation ability.”