The Financial Times reported on October 8 that OpenAI's annualized revenue was approaching $50 billion as of the end of September 2026. That is about $20 billion below the figure of nearly $70 billion widely reported in September. But it would be premature to read this as a sharp drop in OpenAI's sales. The difference comes from how much of the revenue from sales made through cloud providers is counted as OpenAI's own revenue.
The $20 billion gap also matters for comparing OpenAI's growth with that of its rival Anthropic. Once you lay out how revenue is recognized under U.S. accounting standards, and distinguish revenue from investment, it becomes clear why demand for AI services and the financial capacity to support enormous computing infrastructure cannot be judged by the same number.
Why $50 billion and $70 billion differ by $20 billion
According to the FT's report, the $70 billion figure emerged from an adjustment made when OpenAI's investors tried to compare its annualized revenue with Anthropic's on the same basis.
Axios, in its own reporting on October 8, also said the $70 billion figure was OpenAI's revenue converted to a gross basis for the purpose of comparison with Anthropic.
According to Axios, Anthropic books the full amount customers pay for sales made through cloud partners as revenue. OpenAI, by contrast, books only its own share as revenue for some partner-channel sales.
In other words, the two companies account for cloud-channel revenue differently. It does not mean OpenAI makes no sales through cloud providers.
Nor are $50 billion and $70 billion the same calculation applied at different points in time. Calculating a rate of decline from the gap between them would therefore not show an actual change in demand.
It would also be wrong to assume the entire gap of roughly $20 billion represents payments to cloud providers, lower profit, or cash outflow.
Offering AI models through cloud providers has already become a common way to sell them.
For example, in its May 28, 2026 announcement, Anthropic said Claude is available on three cloud platforms: Amazon Web Services (AWS), Google Cloud, and Microsoft Azure.
Contractual relationships and the way payment is received may differ depending on whether a customer contracts directly with an AI developer or uses the model through a cloud service it already relies on. Comparing revenue figures requires checking these differences in sales channels.
How the same $100 transaction can be $100 or $70 in revenue
Consider a transaction in which a customer pays $100 for an AI service. The AI model provider receives $70 of that, and another company involved in the sale receives the remaining $30.
If the transaction is recorded on a gross basis, the model provider books $100 in revenue and records the $30 paid to the other company as an expense.
If it is recorded on a net basis, the provider books only its own $70 share as revenue.
So the same $100 transaction can show up as $100 or $70 in accounting revenue. In this example, however, the amount left before other costs is $70 either way.
This is a hypothetical to illustrate the difference between gross and net presentation. It does not reflect the actual contract terms or margins of OpenAI or Anthropic, and it excludes costs such as the computing needed to run AI model inference.
The key point is that reporting larger revenue is a separate matter from how much profit remains after costs.
Companies also cannot freely choose between gross and net presentation.
The Financial Accounting Standards Board (FASB) explanatory materials on its amendments set out, under the revenue recognition standard Topic 606, the criteria for judging whether a company is a principal or an agent in a transaction.
The central question is whether the company controls the good or service before it is transferred to the customer.
A company that is a principal, supplying the good or service itself, records the consideration received from the customer as gross revenue.
A company that is an agent, arranging for another party to provide the good or service, records as revenue the fee or similar consideration it earns.
Principal and agent are not classifications applied uniformly to a whole company. They are judged by the goods or services provided and the terms of individual contracts, so even the same company may account for revenue differently from one transaction to another.
Accurately comparing the growth of OpenAI and Anthropic therefore requires not just aligning gross and net figures, but also checking which sales transactions are being adjusted.
The reports alone cannot verify the terms of the individual contracts the two companies have signed. Nor can we conclude that the entire $20 billion difference arises from the same type of fee.
An adjusted figure for comparison is useful, but without knowing how it was calculated, it is hard to conclude which company is actually selling more.
Annualized revenue and funding raised are entirely different numbers
In its March 31, 2026 funding announcement, OpenAI said its monthly revenue at the time had reached $2 billion.
Simply multiplying that monthly revenue by 12 gives an annualized $24 billion.
But that is the amount if that month's pace continued for a year; it does not mean the company actually generated $24 billion over the past year.
Annualized revenue is a convenient indicator of the momentum of a fast-growing company. But if usage, pricing, or contract terms change, actual annual revenue can diverge sharply from that figure.
The $50 billion now reported likewise does not represent finalized revenue for a full year, nor income guaranteed by contract going forward.
And when comparing the $24 billion in March with the $50 billion at the end of September, an accurate growth rate cannot be calculated without confirming that the scope of sales transactions covered is the same.
Beyond revenue, figures related to OpenAI's enormous fundraising have also been reported.
The $122 billion the company announced on March 31 is the total capital that investors committed in the funding round. It does not mean the full amount was paid into OpenAI on the day of the announcement.
The $10 billion that SoftBank Group announced on October 1, meanwhile, is the amount of additional investment it actually carried out in OpenAI.
The figures can be sorted as follows.
| Amount | Timing / announcement | What the figure represents |
|---|---|---|
| About $50 billion | October 8 report on the end of September 2026 | OpenAI's annualized revenue |
| About $70 billion | Reported in September; Axios explained the calculation on October 8 | OpenAI's annualized revenue adjusted to a gross basis for comparison with Anthropic |
| $122 billion | OpenAI's March 31 announcement | Total capital committed in the funding round; not all received on that day |
| $10 billion | SoftBank Group's October 1 announcement | Amount actually paid in as the final tranche of a $30 billion additional investment |
Sources: Axios's report on the revenue metric, OpenAI's funding announcement, SoftBank Group's announcement on carrying out the investment.
The $50 billion and $70 billion are annualized figures based on revenue over a given period, while $122 billion is the amount of investment pledged and $10 billion is the amount actually invested. Because the measurement dates and scope also differ, these figures cannot simply be added or subtracted to judge changes in a company's funds.
According to SoftBank Group, the $10 billion payment on October 1 completed the planned $30 billion in additional investment.
With that, its cumulative investment in OpenAI reached $64.6 billion, and its ownership stake reached about 13%. It also said it funded the final $10 billion with proceeds from issuing foreign-currency-denominated senior bonds.
The $10 billion is a figure showing the status of SoftBank Group's investment. It is not revenue from OpenAI's customers, nor is it a separate pool of funds to be added on top of the $122 billion raised.
Revenue reflects customers' use of and payment for services, while investment shows funds supplied by investors to carry out the business.
Both are important for thinking about a company's growth, but they show fundamentally different things. The distinction is essential, especially when judging whether large-scale AI infrastructure investment can be sustained.
Revenue alone doesn't show whether huge AI infrastructure spending can be supported
In its March 31 announcement, OpenAI laid out a strategy of expanding investment in computing capacity, improving its AI models and products to grow its user base, and channeling revenue growth into further investment.
The company said that at the time, enterprise business already accounted for more than 40% of revenue. Alongside the broad adoption of consumer ChatGPT, it aims to turn continuous use of AI in companies' daily operations into revenue growth.
However, even if converting to a gross basis makes the revenue figure larger, customers do not thereby pay any new fees.
Changing how revenue is presented in the accounts also does not increase the cash available to cover the cost of the computing equipment needed to train and run AI models.
To carry out the growth strategy it has set out, OpenAI needs not only to grow demand for its AI services but also to manage the costs of delivering them, secure profitability, and keep raising the investment funds it requires.
That is why comparing the businesses of OpenAI and Anthropic requires combining multiple sources of information, not revenue alone.
First, one should check actual revenue over the same period and understand how far sales through cloud partners are counted as revenue.
Next, checking payments to cloud providers and the computing costs needed to deliver AI models makes it easier to judge how much of the revenue translates into profit.
Further, looking at cash flows from operations and capital spending shows how much of the investment the business can fund from its own cash and how much it depends on outside investment.
The gap between $50 billion and $70 billion is not evidence that demand for OpenAI's AI services has plunged. The two figures were produced by different aggregation methods in the first place.
At the same time, an explanation of the difference in accounting presentation does not prove that OpenAI's business is generating sufficient profit.
In evaluating the growth of AI companies, what matters is not just the revenue figure itself, but how much profit and cash that revenue generates, and whether it can support enormous investment in computing infrastructure.
Once the scope of revenue recognition, the cost structure, and actual cash flows become clear, investors will be able to compare the growth of OpenAI and Anthropic not as a simple contest of revenue scale, but from the standpoint of whether each can sustain its business over the long term.