OpenAI told investors that its revenue was approaching $50 billion on an annualized basis at the end of September, according to an Oct. 8 Reuters report on a Financial Times story. That is roughly $20 billion below the figure of about $70 billion that had circulated in September. In a market where enormous valuations and infrastructure commitments are often justified by explosive growth, that difference matters — but not necessarily for the reason the headline suggests.
The real question is not whether OpenAI suddenly lost $20 billion of business. On the reporting now in the public domain, there is no basis for that claim. The more useful question is whether AI companies’ top-line growth figures are being presented on comparable enough terms to support big conclusions about demand, durability and capital needs.
So far, this looks less like a sales collapse than a measurement problem.
A run rate is not the same as revenue
The first complication is the number itself. An annualized revenue figure, or run rate, is a projection from a recent period over a full year. It is not the same thing as recognized revenue for a completed fiscal year, cash collected, deferred revenue, gross margin or operating profit. It can move sharply depending on the month chosen, the mix of customers and products, how much business flows through partners, and how usage-heavy a period was.
That distinction matters because the OpenAI figures in circulation do not come from public audited financial statements. Reuters said the Financial Times report was based on financial documents shared with investors, that OpenAI did not immediately respond to a request for comment, and that Reuters could not independently verify the report.
That leaves a narrower, but still important, conclusion: investor-facing materials appear to have described a roughly $50 billion annualized business at the end of September, while a higher figure of around $70 billion had previously been signaled in media and investor discussions. The gap is real as a reporting issue. What it means operationally is the harder question.
The likely source of the gap: cloud-partner sales
The most persuasive explanation so far is that OpenAI and Anthropic have been presenting some cloud-partner sales differently. Axios independently reported the same broad discrepancy and said Anthropic includes the full customer payment from certain cloud-partner sales in its top-line figure, then records the partner’s share as an expense. OpenAI, by contrast, records only its own share of some partner-routed sales.
That may sound technical, but it can change the headline number dramatically.
Imagine a customer buying AI services through a cloud platform such as AWS or Google Cloud. If the cloud provider bills the customer and keeps part of the payment, one company may count the full customer payment as revenue and treat the partner’s cut as a cost. Another may count only the portion it expects to keep. The customer may pay the same amount either way. The model provider may ultimately receive the same economic share either way. But the top line can look much bigger under the first method.
This is the classic principal-versus-agent problem in modern software distribution: who controls the customer relationship, who sets pricing, who bears the delivery obligation, and who is really making the sale? Different answers can support different revenue presentation. Axios reported that both approaches can be GAAP-compliant.
That is why the reported $50 billion versus $70 billion gap should not be read as proof that $20 billion vanished. A lower number may reflect a more conservative, net presentation rather than weaker demand. A higher number may be useful for understanding total customer spend flowing through an ecosystem, but less useful for ranking companies side by side.
Why this matters beyond a single headline
The AI boom has been financed on the assumption that leading model companies are scaling fast enough to justify massive outlays on chips, cloud capacity and data centers. When the headline metric behind that story turns out to depend heavily on presentation, the issue is bigger than one company’s investor deck.
For investors, the takeaway is straightforward: a run rate on its own is not enough. If AI companies route meaningful sales through cloud partners, then gross and net presentations can make one provider look larger or faster-growing than another without changing underlying customer demand. That makes annualized top-line figures a poor league table unless the basis is clearly stated.
What matters more is the set of questions behind the number. How much revenue has actually been recognized over a completed period? How much cash has been collected? What share of sales runs through partners? What is gross margin after inference and cloud costs? How concentrated is the customer base? Are usage levels sticky, or driven by temporary bursts and discounts? And how large, fixed and cancellable are the compute commitments signed to support that growth?
Those questions matter just as much for suppliers and enterprise buyers. Chipmakers, cloud providers and data-center operators use growth signals from frontier AI labs to plan capacity and forecast demand. If the market leans too heavily on non-comparable top-line metrics, suppliers risk overreading demand visibility. Enterprise customers face a different concern: not whether one revenue headline is bigger, but whether their vendor’s usage is durable enough and margins healthy enough to support product road maps, service levels and long-term pricing.
None of this makes annualized figures useless. In fast-growing private markets, they are a shorthand for momentum. They can capture a business accelerating too quickly for last year’s reported revenue to tell the story. But shorthand becomes risky when it is treated as a precise measure of scale, especially in a sector where distribution channels and compute economics can reshape the headline.
The missing piece is still public financial disclosure. The record does not show OpenAI’s recognized revenue by period, the exact contract terms behind the reported figures, the share of sales routed through specific cloud partners, or how much of the gap between $50 billion and $70 billion reflects presentation versus timing, mix or one-off items. It also does not show gross margin, cash burn or the cost of serving those sales.
Until those details exist in public statements, the smarter reading is not that OpenAI missed a confirmed target or exposed a hidden hole in demand. It is that AI’s most eye-catching growth numbers can be technically defensible and still fail the comparability test that investors, suppliers and enterprise customers increasingly need.
By
By
By
By
By
