Last week, OpenAI told investors its annualized revenue was approaching $50 billion. The week before, the same company was reportedly on track for $70 billion. That $20 billion gap didn’t come from lost customers or a demand collapse. It came from an accounting disagreement about what counts as revenue — and it says more about the AI industry’s economics than any earnings report.
The Financial Times broke the story on October 8, citing documents shared with investors. The $70 billion figure, widely reported by Axios and others in September, was devised by OpenAI’s own investors trying to make the company’s numbers comparable to Anthropic’s. The problem: Anthropic counts the full value of sales through cloud partners like AWS, Google Cloud, and Microsoft Azure as revenue. OpenAI doesn’t.
So when investors “grossed up” OpenAI’s numbers to match Anthropic’s methodology, they got $70 billion. When OpenAI reported its own figures, it got $50 billion. Both numbers are technically correct. They’re just measuring different things.
The Accounting Gap
Here’s how the two companies differ. If a customer pays $100 for an AI service through Amazon Bedrock, Anthropic records the full $100 as top-line revenue and lists Amazon’s cut as an expense. OpenAI records only its share of certain partner sales. Both approaches are GAAP compliant. They produce very different headline numbers.
Anthropic’s annualized revenue crossed $65 billion in July and is expected to reach $100 billion by year-end, according to Reuters. OpenAI’s $50 billion is impressive on its own — but it’s not $100 billion. The race between the two labs, as I covered in the OpenAI-Anthropic race, is far from over. The gap isn’t just accounting. It’s distribution. Anthropic’s models are available natively across AWS, Google Cloud, and Azure, giving it reach into hundreds of millions of enterprise seats. OpenAI’s partnership with Microsoft is deeper but narrower. — and pricing pressure from open-weight models is only adding to the pressure
The market reaction was swift. The Nasdaq fell 1.25%, its worst day since mid-August. Oracle dropped 5.5%, Nvidia lost 2.9%, and AMD fell 3.3%. The S&P 500 Information Technology sector declined 1.8%, making it the worst-performing sector of the day. Investors weren’t just reacting to a number. They were reacting to the realization that the AI trade might be built on shakier foundations than the headlines suggested.
What the Numbers Actually Say
Let’s put this in context. OpenAI’s 2025 financials, leaked earlier this year and verified by the FT, showed $13.07 billion in revenue against $34 billion in total costs. The operating loss was $20.92 billion. The net loss, inflated by a one-off accounting charge related to OpenAI’s conversion to a for-profit entity, was $38.5 billion.
Those are staggering numbers. But the trajectory matters more than the absolute figures. Revenue nearly quadrupled from $3.7 billion in 2024 to $13.07 billion in 2025. The expense ratio improved from $2.37 for every dollar of revenue to $1.60. OpenAI is moving toward profitability, but it’s not there yet.
The $50 billion annualized figure represents a massive acceleration from $13 billion in 2025. Even if it’s $20 billion less than the investor-adjusted number, it’s still a 4x year-over-year growth rate. The question isn’t whether OpenAI is growing. It’s whether the growth rate justifies the spending.
The Microsoft Factor
One detail from the 2025 financials stands out: OpenAI paid Microsoft $17.2 billion last year. That includes $10.59 billion in R&D expenses (primarily model training) and $6.047 billion in cost-of-revenue charges (cloud computing). Microsoft is simultaneously OpenAI’s biggest partner and its biggest expense.
This creates a strange dynamic. Microsoft owns 49% of OpenAI’s for-profit entity. It also bills OpenAI billions for compute. The partnership is symbiotic, but it’s not simple. When OpenAI’s revenue is reported, the Microsoft relationship complicates the picture. Some of the $20 billion “gap” between the $70 billion and $50 billion figures is essentially the value of sales flowing through Microsoft Azure — sales that OpenAI doesn’t count as its own revenue but that Microsoft certainly counts as its own.
As I noted when Microsoft and OpenAI are renegotiating their partnership, the partnership is shifting from exclusive to strategic. Microsoft is building its own AI capabilities while remaining OpenAI’s primary cloud provider. The revenue accounting discrepancy is a symptom of that complexity.
What This Means for the AI Trade
The immediate takeaway is that AI revenue numbers are messy. Annualized run rates are inherently imprecise — they extrapolate a single month’s sales across a full year. When you layer different accounting methodologies on top of that, you get numbers that look comparable but aren’t.
As I wrote in my piece on the broader AI economics question, the deeper takeaway is that the AI industry’s economics are still unsettled. OpenAI and Anthropic are both growing rapidly, but neither is profitable. The infrastructure spending — data centers, chips, talent — is happening now. The revenue is catching up, but the gap between spending and revenue remains enormous.
For companies evaluating AI tools, this is a reminder to look beyond the headline numbers. A vendor claiming $70 billion in annualized revenue might be using a generous accounting method. That doesn’t make the product bad, but it should make you ask harder questions about unit economics, pricing models, and the total cost of ownership — something I explored when Satya Nadella warned companies about feeding secrets to their future competitors.
The Bigger Picture
OpenAI is expected to go public next year, with Anthropic potentially IPOing as soon as November. When that happens, both companies will have to report standardized financials. The accounting discrepancies will disappear. Investors will see clean, comparable numbers — and they may not like what they see.
The AI industry is at an inflection point. The technology is real, the demand is real, and the growth is real. But the economics are still catching up. The $20 billion revenue gap isn’t a scandal. It’s a sign that the industry is maturing — that the numbers are starting to matter more than the narratives.
As someone who manages IT budgets and evaluates AI tools for government use, I’ve learned to be skeptical of headline revenue figures. The real question isn’t how much money a company says it’s making. It’s how much value it’s creating for customers — and whether the unit economics work at scale.
OpenAI’s $50 billion is a remarkable achievement. It’s also a reminder that in AI, the numbers are never as simple as they look.