New data from OpenAI indicate an emerging split among global companies between those that adopt AI aggressively and those that move more slowly. According to the report, firms that implement AI widely are pulling ahead of peers that have been more cautious.
Concrete figures
OpenAI’s report shows that in June the top 10% of companies by AI usage consumed on average 8.3 times as many output tokens per user as companies in the middle of the pack. That represents a sharp increase compared with January, when the top firms used 2.6 times the median. The report was shared first with Semafor.
Shift from ChatGPT to Codex
The study also finds that OpenAI’s business customers as a group are now using more tokens on Codex — the company’s agentic coding platform — than on the traditional ChatGPT interface. This suggests businesses are adopting AI for tasks that involve more automated execution rather than simply asking questions. Ronnie Chatterji, OpenAI’s chief economist, described to Semafor that companies are shifting toward "work that’s more agentic rather than asking."
Competitive and business implications
OpenAI’s push to win more corporate customers is partly aimed at catching up with rival Anthropic, which focused early on businesses and coding tools in addition to the consumer market. Both OpenAI and Anthropic are preparing for major initial public offerings (IPOs).
Observers note the pattern echoes past technological disruptions such as the spread of the internet, when early adopters pulled ahead and others caught up once productivity gains were demonstrated. Chatterji said the central challenge for CEOs and CTOs is to convince boards and shareholders that rising AI expenditures are justified.
Measuring ROI remains difficult
Chatterji also warned that high token usage does not necessarily translate directly into productivity, and that accurately measuring return on investment remains "frankly difficult to do." The divergence in usage metrics therefore highlights a growing adoption gap, while the extent of real business benefit and how to quantify it stay uncertain.
