If your company pays for ChatGPT or Claude, the figure circulating about what each provider generates may depend on accounting criteria rather than how many customers they have. OpenAI reached an annualized revenue run rate of about $50 billion at the end of September, according to the Financial Times, while it negotiates a new capital round of at least $30 billion.
The data matters now because it coincides with a delicate moment for the generative artificial intelligence market. A report on these figures triggered drops in tech and chip stocks, and puts into perspective the IPO plans for both OpenAI and its main rival, Anthropic.
Why the same figure appears as $50 billion and $70 billion
At the end of September, a figure of nearly $70 billion circulated after Axios reported that OpenAI was approaching that level of revenue. However, the Financial Times placed it at around $50 billion at the end of the month.
The difference does not reflect a real cut in sales. According to Axios, the higher figure was calculated to make OpenAI's revenue more comparable to Anthropic's, as each company accounts for sales made through commercial partners differently.
The accounting criterion that makes the difference
Anthropic records the full customer payment as revenue when selling through cloud providers, and the portion the provider keeps is later recorded as an expense. OpenAI, on the other hand, in certain agreements with partners, only counts the portion it effectively retains as its own revenue.
Both methods comply with US accounting standards known as US GAAP, according to Axios. The choice depends on who controls the relationship with the end customer and who is responsible for product delivery, not on any manipulation of figures.
Stock market drop after the Financial Times report
Following the publication of the lower figures, tech stocks fell and chip manufacturer stocks lost several percentage points. The episode confirms the market's sensitivity to any data regarding the two largest generative AI companies in the United States.
That reaction may also explain the caution of Anthropic, which was preparing its IPO for November. OpenAI, for its part, had already postponed its own to next year. Its CEO, Sam Altman, attributed the delay to security risks, although the postponement had been decided before the cybersecurity incidents recorded in recent months. In April, it had already been reported that the company missed its internal growth targets.
The $70 billion forecast for the end of the year
Bloomberg reports that OpenAI expects to reach an annualized revenue run rate of at least $70 billion by the end of 2026. The company shared this projection during its new funding round talks, so it is an internal estimate and not a settled fact.
The main momentum comes from the business segment. According to CNBC, OpenAI's total annualized revenue grew 77 percent in the third quarter, while revenue from the enterprise segment rose 107 percent in the same period.
The $30 billion sought and the $1.4 trillion valuation
OpenAI is currently negotiating at least $30 billion in new capital, with a target pre-money valuation of $1.4 trillion. In March, it had raised up to $122 billion with a post-money valuation of $852 billion.
The underlying debate remains open: whether the revenue of these companies can grow at the same pace as their enormous spending commitments on data centers and computing capacity. The answer will depend on the productivity improvements that corporate clients manage to demonstrate over time.
Where the demand comes from: ChatGPT, Codex, and GPT-6.1-Sol
During its DevDay event, OpenAI announced that ChatGPT exceeds 1.2 billion weekly users. More than 35 million people use ChatGPT Work and Codex every week, and 2.5 million companies use its products, according to data from the company itself.
Growth is supported, according to Axios, by sales aimed at companies and a price war against Claude and models developed in China. OpenAI reinforced that strategy with the launch of GPT-6.1-Sol. Codex, its programming assistant, is also gaining popularity alongside the GPT-6 model family.
What it means when comparing OpenAI with Anthropic
For anyone comparing AI provider revenue for purchasing or investment purposes, it is advisable to first review how each one accounts for sales made through partners. With figures of $50 billion versus nearly $70 billion for the same company, a direct comparison with Anthropic can be misleading.
Those negotiating artificial intelligence contracts can benefit from the price war described by Axios. And for those following the markets, it is worth remembering that a single report was enough to move chip stocks by several points. OpenAI's IPO was postponed to next year, while Anthropic's was being prepared for November.
Frequently Asked Questions
How much does OpenAI actually generate per year?
According to the Financial Times, OpenAI reached an annualized revenue run rate of about $50 billion at the end of September 2026. Other estimates, calculated using a different accounting criterion, placed it near $70 billion.
Why are there two different figures for the same revenue?
The difference is due to how sales made through commercial partners are accounted for. One method records only the portion of revenue that OpenAI retains; another, calculated to compare with Anthropic, includes the total customer payment.
How much money is OpenAI looking to raise now?
The company is negotiating at least $30 billion in new capital, with a target pre-money valuation of $1.4 trillion.
When will OpenAI go public?
OpenAI postponed its IPO to next year. Its CEO, Sam Altman, cited security risks, although the delay had already been decided before the recent cybersecurity incidents.
How many users does ChatGPT currently have?
According to data shared by OpenAI at its DevDay event, ChatGPT exceeds 1.2 billion weekly users, and 2.5 million companies use its products.
How does this affect the comparison with Anthropic?
Anthropic records the full customer payment on sales through cloud partners, while OpenAI only records its share in certain agreements. Both methods are valid under US accounting standards, but they generate revenue figures that are difficult to compare directly.
