OpenAI says ChatGPT now reaches 1.2 billion people every week

At DevDay, OpenAI updated its usage scale: ChatGPT reaches more than 1.2 billion users weekly; ChatGPT Work and Codex together exceed 35 million weekly active users; and 2.5 million businesses already use its products. These three sets of numbers cover three levels — the consumer side, developers and office scenarios, and enterprise customers — forming the basic picture of the growth OpenAI is presenting to the outside world.

It should be noted that these numbers all come from OpenAI's announcement at DevDay. The sources do not provide independent third-party verification, nor do they disclose the statistical basis or deduplication method for the 1.2 billion weekly active users. They should therefore be understood as company-reported scale metrics, not externally audited results.

According to Axios, which first reported it, OpenAI's annualized revenue run rate is approaching $70 billion and has grown about 70% since the start of the third quarter. The annualized revenue run rate (ARR) metric works by converting current monthly revenue into a full-year figure, so it is an extrapolated estimate and is not equivalent to revenue actually realized.

Putting these two numbers together: a scale approaching $70 billion, combined with intra-quarter growth of about 70%, shows that OpenAI's current revenue curve is quite steep. But the sources do not give the absolute base for this growth rate, the cutoff date for the statistics, or the revenue composition, so it is impossible to judge from this whether the growth rate can continue.

The sources attribute OpenAI's growth to three threads: enterprise sales, the Codex coding assistant, and an aggressive price war against Claude and Chinese models. Codex is described as growing rapidly on the momentum of the GPT-6 model family, which is consistent in direction with the data showing ChatGPT Work and Codex together exceeding 35 million weekly active users.

OpenAI has just launched GPT-6.1-Sol, and the sources say the company is doubling down on its price war strategy with it. However, the sources do not explain the specific capabilities or release timing of GPT-6.1-Sol and the GPT-6 model family, nor do they give figures for the scale of the price war, so the idea that "the new model drives growth" is currently only the sources' narrative and lacks data support.

The sources provide a comparison: Anthropic's annualized revenue run rate reportedly exceeded $65 billion in July and may now have caught up with or surpassed OpenAI; at the same time, Anthropic is preparing to launch an IPO as early as November. If this comparison holds, OpenAI's annualized revenue run rate approaching $70 billion does not constitute a clear lead.

But both pieces of information are secondhand accounts. The sources use the phrase "reportedly," and "may have caught up with or surpassed" is itself speculative. Anthropic's actual current revenue level, and whether the November IPO can launch on schedule, cannot be confirmed from the available sources.

The sources put the core question directly: can revenue grow fast enough to cover massive data center bills. This is a question about matching income and expenses — user scale and annualized revenue run rate answer "how fast money is coming in," while data center spending answers "how large the outgo is."

The problem is that the sources give neither the specific scale of data center spending nor the magnitude of the price war's costs. Without these two figures, it is impossible to judge whether an annualized revenue run rate approaching $70 billion is comfortable or tight relative to the bills. Therefore, "can revenue cover costs" can currently only remain an open question to watch, not a conclusion that can be drawn.

What can be relatively certain is the definition of the ARR metric: converting current monthly revenue into a full-year figure. Beyond that, almost all the key numbers in this article — 1.2 billion weekly users, 35 million weekly Work and Codex users, 2.5 million businesses, an annualized revenue run rate approaching $70 billion, and about 70% intra-quarter growth — come from OpenAI's DevDay announcement or Axios reporting. They are source claims and lack independent third-party verification.

The Anthropic-related revenue and IPO information is one level weaker still, being secondhand accounts. As for "can revenue cover data center bills," the sources themselves already frame it as a question, and this article likewise keeps it as an unresolved issue. When citing these numbers, readers should distinguish among three types of information with different levels of strength: company self-reports, media reports, and speculation.