A new figure is reshaping how Microsoft's AI revenue looks on paper. On August 5, Bloomberg estimated that roughly 65-70% of the annual run rate Microsoft has disclosed for its AI business traces back to OpenAI. OpenAI is a Microsoft cloud customer, and revenue-sharing arrangements remain part of the two companies' agreement. This is a ratio worth taking seriously when thinking about how Microsoft's AI revenue is actually constructed.
But this roughly 70% figure is not a customer-level accounting line item that Microsoft itself disclosed in its financial statements. It is Bloomberg's estimate, and the denominator is neither Microsoft's total company revenue nor its entire Microsoft Cloud business. The narrow AI-business run rate and the broader cloud-business results need to be read separately.
What denominator does the 70% estimate use?
Microsoft stated in FY26 Q3 that its AI business's annual revenue run rate exceeded $37 billion, up 123% year-over-year. This is a metric that annualizes the business's scale at a given point in time—it is not the sum of GAAP revenue recognized during FY26. Even in its Q4 earnings materials, Microsoft has not disclosed AI revenue as a standalone accounting segment or as customer-specific revenue.
Microsoft posted $90.07 billion in revenue for FY26 Q4 and $331.839 billion for the full fiscal year. The $37 billion-plus AI run rate is not revenue recognized on top of these figures during the same period—it's an expression of the AI business's current annualized scale. Not conflating actual accounting figures with business metrics is the starting point for correctly understanding the scope of that roughly 70% figure.
Bloomberg's approximately 70% estimate places this $37 billion-plus AI run rate in the denominator. It therefore does not mean "roughly 70% of Microsoft's revenue comes from OpenAI." Microsoft's total FY26 company revenue was $331.839 billion, which is an actual figure recognized over the course of the fiscal year. Treating an annualized run-rate metric and an annual actual figure as the same kind of ratio breaks down the basis for comparison.
For reference, if you were to line up the $37 billion against the full-year FY26 figure of $331.839 billion, you'd get about 11.1%. However, the former is an annualized run rate and the latter is an actual annual figure, so this is not a precise share calculation. The details of Bloomberg's methodology, including how it connects OpenAI-related amounts to the $37 billion-plus figure, cannot be confirmed from Microsoft's public disclosures. The roughly 70% figure should be understood not as an audited disclosure from Microsoft, but as Bloomberg's analysis based on the company's disclosures.
The picture changes with the broader cloud business
Microsoft Cloud posted $214.4 billion in revenue for FY26. Q4 alone reached $59.3 billion, a broad metric that includes Microsoft 365 Commercial cloud, Azure and other cloud services, commercial LinkedIn, and Dynamics 365. Microsoft has stated that roughly 90% of this annual Microsoft Cloud revenue came from customers other than frontier model companies.
This 90% figure is not a breakdown of the $37 billion-plus AI run rate. Conversely, Bloomberg's roughly 70% figure does not represent OpenAI's share of Microsoft Cloud as a whole. Both can be true at once: within the narrow denominator of the AI business, concentration on OpenAI is significant, while within the broader denominator of Microsoft Cloud, the customer base is diversified.
Metrics indicating future revenue also cannot be substituted for actual revenue. Microsoft's commercial RPO (remaining performance obligation) stood at $678 billion, up 84% year-over-year. The weighted average duration of RPO, including OpenAI, is 2.3 years, with roughly 30% expected to be recognized as revenue over the next 12 months. RPO represents contracted performance obligations—it is not revenue or profit already recognized.
Q4 commercial bookings growth, year-over-year, was 18% excluding OpenAI, and 10% (11% in constant currency) including OpenAI's Azure commitments. Microsoft also stated that all of the net increase in commercial RPO during the quarter came from customers other than frontier model companies, with RPO excluding OpenAI growing 25%. These figures indicate how contracts are growing, but they do not directly establish what percentage of AI revenue OpenAI accounts for.
The Azure-centered agreement continues
On April 27, Microsoft and OpenAI announced a revised partnership agreement. Microsoft remains OpenAI's primary cloud partner, and OpenAI's products will be offered on Azure first, except in cases where Microsoft cannot or does not provide the necessary capacity. OpenAI, meanwhile, can offer all of its products on any cloud provider. The tie to Azure remains strong, but this is not a contract that permanently and exclusively locks in all workloads.
The economic connection isn't one-directional either. Under the revised agreement, Microsoft's revenue-sharing payments to OpenAI will end, while OpenAI's revenue-sharing payments to Microsoft will continue at the same rate through 2030, subject to a cap. Microsoft's license to OpenAI's intellectual property continues through 2032, but on a non-exclusive basis.
There are two revenue pathways within this agreement: cloud revenue generated from OpenAI's use of Azure compute resources, and the revenue-sharing arrangement. When reading Bloomberg's estimate, one cannot simply layer revenue from AI-enabled Microsoft products together with infrastructure usage and contractual income tied to OpenAI as if they were the same customer-level revenue figure.
Can the AI revenue breakdown be disclosed?
In its Q4 earnings discussion, Microsoft stated that it offers over 11,000 models, combining third-party models including OpenAI's with its own MAI family. Using a system that makes models interchangeable, Maia 200 reportedly supports both OpenAI models and MAI models. Analyses estimating revenue concentration on OpenAI and the breadth of models Microsoft offers illuminate the company's AI business from different angles.
There's also material for growth on the proprietary product side. Microsoft 365 Copilot has about 30 million paid seats, and GitHub Copilot revenue grew more than 60% quarter-over-quarter after the introduction of usage-based billing. However, Microsoft has not disclosed a total AI revenue figure broken down by customer or product that would include these. Seat counts and growth rates alone don't make it possible to calculate how far non-OpenAI AI revenue has grown.
Q4 capital expenditures were $41 billion, roughly two-thirds of which went primarily toward shorter-lived assets like CPUs and GPUs. Who uses this large-scale compute infrastructure, and how much, will shape the outlook for return on investment. Even so, Bloomberg's roughly 70% estimate cannot be used to draw conclusions about Microsoft's profit margins or the level of non-OpenAI demand.
What should be verified next is how far Microsoft will go in disclosing a customer- and product-level breakdown of its AI business. Keeping the three denominators separate—the $37 billion-plus AI run rate, the $214.4 billion in FY26 Microsoft Cloud revenue, and the $331.839 billion in total FY26 company revenue—and tracking Bloomberg's 65-70% estimate as applying to the first of these, is the condition for not misjudging how dependency is actually changing.