AI adopter gap widens: top firms now use 8.3× more tokens than the median

OpenAI's latest internal data provides a clear picture of diverging adoption patterns among its business customers. The gap between the top 10% of AI users and the median company has expanded dramatically in a six-month window. In January, firms in the top decile utilized 2.6 times the median token consumption per user. By June, that ratio had ballooned to 8.3×, indicating a sharp acceleration in the pace at which leading adopters are outpacing the rest of the enterprise market. The data was shared first with Semafor and reflects consumption across OpenAI's entire business customer base. The shift in which tools companies are using offers insight into what this acceleration means in practice. Collectively, OpenAI's business customers are now consuming more tokens on Codex, the company's agentic coding platform, than on the traditional ChatGPT interface. This pivot is significant because it signals a move away from interactive, human-directed AI queries toward AI systems designed to operate more autonomously. According to Ronnie Chatterji, OpenAI's chief economist, businesses are embracing AI for "work that's more doing rather than asking." In other words, rather than using AI to answer questions or draft content on demand, enterprises are building AI into their operations to execute tasks independently and continuously. This distinction helps explain both the widening gap and its rapid acceleration: early adopters are not just using AI more; they are embedding it deeper into their production systems, where it generates far more tokens through continuous autonomous operation than traditional question-and-answer interfaces ever could.