Two AI giants are taking different main paths. Barclays: OpenAI leads consumer-level AI, while Anthropic wins the favor of enterprise customers.

OpenAI and Google's latest disclosed data show that in the battle for ordinary consumers, ChatGPT has already left the search giant Google far behind, and the gap is widening. At a recent developer day, OpenAI announced that its API is consuming about an astonishing 60 billion tokens per minute, a twenty-fold increase from two years ago; at the same time, ChatGPT's weekly active users surpassed 800 million. In contrast, Google's numbers are more macroscopic: all of its services AI Overview, Gemini App, Google Cloud API, etc. collectively process 13 trillion tokens per month. It is understood that tokens the smallest unit of calculation consisting of words, roots, and even punctuation determine the cost and revenue tables of the large model world, as well as determine how much of a "free card" manufacturers can play in front of consumers. After piecing together these fragments, Barclays analyst Ross Sandler estimates that although Google is far ahead in terms of "total volume," when looking at the "native consumer AI applications" sub-track, OpenAI's token consumption is still more than twice that of Gemini. In other words, the 900 trillion tokens brought in by search users helped Google save face, but also exposed the reality: the product that truly gets consumers to actively open up and chat, ChatGPT is the one that is more "burning money" and "attracting money". Sandler stated that with 800 million weekly active users, ChatGPT can generate over 4.2 quadrillion inference tokens per month, and if estimated based on a possible 1.2 billion monthly users, this number will be even higher; whereas Gemini App has a monthly active user base of 450 million, producing only about 1.9 quadrillion tokens, which is less than half of the former. In the second quarter earnings conference call, Google revealed that Gemini App's daily request volume increased by 50% month-on-month, but no matter how fast the requests increase, they cannot keep up with the "token inflation" pace of ChatGPT. However, the competitive landscape in the enterprise market has shown a different trend: with the title of "OpenAI's strongest rival", Anthropic has captured "almost" double the revenue share of OpenAI in the enterprise market with the launch of Claude in March 2023. Internal targets show that Anthropic's annualized revenue run rate will reach $9 billion by the end of 2025 and then increase to $20 billion by the end of 2026; as of August 2025, its run rate has already exceeded $5 billion, with over 300,000 enterprise customers and nearly seven times more customers paying over $100,000 annually. Sandler bluntly stated that Anthropic has built up a "considerable lead" in the new track of "enterprise inference tokens". Returning to Google's timeline: the casual remark in the second quarter conference call that "search users have exceeded 2 billion, now closer to 2.5 billion" explains where the 900 trillion tokens come from and once again reminds the outside world that search is still Google's largest "token engine". While OpenAI holds the consumer-level entrance with 800 million weekly actives, and Anthropic captures B-end budgets with 300,000 enterprise customers, the two lines on the left and right are squeezing Google in the middle. Google still relies on search for half of its tokens but must also accelerate its pace to prove that it is not just a "question and answer cash machine".