This story is not about someone else’s billions. It’s about the fact that behind every convenient AI interface there is always infrastructure, contracts, dependencies, and compute bills. If even Microsoft and OpenAI, two giants of the AI market, are still this financially intertwined, small businesses definitely shouldn’t treat choosing an AI tool like signing up for a regular monthly subscription.

According to TechCrunch AI, Microsoft said in a regulatory filing that it received $10 billion in revenue from commercial arrangements with OpenAI in the quarter ended September 30. That figure includes revenue-sharing payments. It amounts to nearly 11% of Microsoft’s quarterly revenue of $92.3 billion.

The number itself is impressive. But what’s more interesting is this: the companies are already competing for some of the same customers, yet financially they still remain tightly linked.

There is always a cloud bill behind an AI product

For the full 2026 fiscal year, Microsoft’s revenue from OpenAI reached $24.1 billion. At the pace of the latest quarter, that could imply roughly $40 billion a year from OpenAI-related arrangements. As of the end of September, Microsoft also had $7.2 billion in OpenAI receivables. At the end of June, it was $6 billion. This is money from past transactions, not new revenue.

There is no exact public breakdown between revenue sharing and Azure payments. TechCrunch AI separately says it requested more details. Earlier, the publication reported, citing leaked documents, that Microsoft received $865.8 million in net payments from OpenAI for revenue sharing in the first nine months of 2025.

According to reports, OpenAI shares 20% of its revenue with Microsoft under a previous agreement tied to Microsoft’s more than $13 billion investment. The companies have not publicly confirmed that percentage. In April 2026, the two sides agreed that OpenAI would continue sharing revenue with Microsoft through 2030, with certain limits, even if it works with other cloud providers.

And that’s where things get most relevant for companies launching AI agents, RAG search over documents, or support automation. The cost of such a solution does not end with the model price. There is hosting, vector databases, integrations, API limits, data storage, SLAs, and legal risk. The provider changes the terms, and that quickly shows up in your P&L.

AI should be implemented with an exit plan

Imagine an online store that connected an AI agent to handle chat inquiries. For the first two weeks, everything looks great: less manual work, faster replies, managers can finally breathe. Then traffic grows, the number of model requests increases, translation is added, product descriptions are generated, and the knowledge base is searched. Suddenly AI is no longer an experiment — it’s a normal operating expense.

We’d approach it pragmatically:

  • start with one process where it’s easy to measure time or cost savings;
  • document which models, clouds, and services are being used so you don’t lose control of the architecture;
  • don’t keep critical logic only inside one closed tool if it will be hard to move later;
  • build in a scenario for replacing the model or provider at the MVP stage already.

For a clinic, that might be an AI assistant for initial patient routing. For an agency, it could be automating briefs and first drafts of proposals. For a delivery service, it may be analyzing customer requests and the common reasons for delays. The situations are different, but the question is the same: what happens if the price, limits, or availability of the service changes?

Microsoft and OpenAI have been gradually loosening the exclusivity of their partnership since October 2025. Microsoft gave OpenAI more freedom to work with other cloud providers. OpenAI, in turn, committed to buying an additional $250 billion in Azure services. Microsoft’s stake in OpenAI also shrank: from about 25% in June to 24% in September, after around 27% following recapitalization in October 2025.

AI has already become standard business infrastructure. Like CRM, telephony, or a payment service. It should be adopted, but with a cool head: calculate unit economics, don’t confuse a demo with production, and build automation so the business doesn’t become hostage to a single vendor.