For the first time, a single quarterly earnings report from one of the world’s largest cloud providers has publicly quantified what it costs to back two competing frontier AI labs at the same time you are building your own. On July 29, 2026, Microsoft reported Q4 FY2026 revenue of $90.0 billion and net income of $35.8 billion, and buried two numbers in the footnotes that explain the strategic shift the company has been telegraphing for a year: a $3.2 billion gain on its Anthropic investment and a $480 million OpenAI-related gain in the same quarter (Microsoft Investor Relations - FY26 Q4 Press Release).
For readers who buy AI through Azure, Copilot, or one of Microsoft’s agent platforms, the news is not just that Microsoft is doing well. The shape of the business underneath those numbers is changing: Microsoft is now simultaneously a financial backer of OpenAI, a financial backer of Anthropic, and a builder of competing in-house MAI models, while telling enterprise customers that any model in the stack should be swappable at any moment.
The three-way model bet, in numbers
Microsoft’s stake in OpenAI is the most familiar. The company owns roughly 27 percent of OpenAI and hosts its models through a multi-year cloud partnership that has long been the foundation of the Copilot product line. In Q4 FY2026, OpenAI-related gains added $480 million to net income and $0.07 to diluted EPS, and the full fiscal year contribution from OpenAI was $4.96 billion to net income and $0.67 to diluted EPS (Microsoft Investor Relations - FY26 Q4 Press Release).
The Anthropic relationship is newer and faster. Microsoft invested $5 billion in Anthropic in November 2025 as part of a deal in which Anthropic committed to $30 billion in Azure services. The Q4 FY2026 read-through: a $3.2 billion gain on the Anthropic stake, contributing roughly $0.27 to diluted EPS (TechCrunch - Microsoft logs $3.2B from Anthropic investment, but OpenAI was a mixed bag).
The third leg is the one that does not show up in the equity disclosures. Microsoft AI, the in-house group led by Mustafa Suleyman, has been pushing the MAI model family through Microsoft Foundry. On April 2, 2026, Microsoft shipped three MAI models at once: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 (Microsoft AI - Announcing 3 new world class MAI models, available in Foundry). The company has also put MAI-Thinking-1 into private preview, a reasoning model with 35 billion active parameters and roughly 1 trillion total parameters in a Mixture-of-Experts architecture, which Microsoft is positioning as a mid-sized alternative to OpenAI’s o-series and Anthropic’s high-end models (Microsoft AI - MAI-Thinking-1 model page).
Put together, the same customer is now buying from a company that gets paid no matter which frontier lab they choose. The split between labs that sell subscriptions, labs that sell ads, and labs that sell compute keeps shaping the choice, and we traced that arc in our look at the AI business model split earlier this year.
Why Nadella is telling customers to swap models
On the same call, CEO Satya Nadella made the argument explicit. The customer should be in control of its own model choice, and any model at any given time should be swappable (TechCrunch - Microsoft is openly competing with OpenAI and Anthropic more than ever). The framing is not new for hyperscaler marketing, but Microsoft is now in a position to mean it: its catalog includes OpenAI, Anthropic, Mistral, xAI, and its own MAI family, plus everything developers can pull from Hugging Face endpoints.
The pricing signals reinforce the message. MAI-Voice-1 generates 60 seconds of audio in one second and starts at $22 per 1 million characters. MAI-Image-2 starts at $5 per 1 million tokens of text input and $33 per 1 million tokens of image output. MAI-Transcribe-1 supports 25 languages and starts at $0.36 per hour. Microsoft says these models are cheaper than competing offerings (Microsoft AI - Announcing 3 new world class MAI models, available in Foundry).
Microsoft’s reach at the application layer is also doing more work than the model list alone. The company reported that Microsoft 365 Copilot has crossed 30 million paid seats and that Azure passed $100 billion in annual revenue for the first time in FY2026 (Microsoft Investor Relations - FY26 Q4 Press Release). Copilot is the harness layer that consumes whatever model is on top, and at this scale Microsoft can route revenue across OpenAI, Anthropic, and MAI models without losing the customer relationship. We covered the explicit push to put Anthropic’s reasoning inside that harness when Copilot Cowork launched as an enterprise agent earlier this year.
What This Means
For enterprise customers, the practical effect is that the answer to “which model does Copilot use?” is becoming an architectural choice rather than a vendor choice. The same Azure bill can fund OpenAI fine-tuning, Anthropic reasoning calls, and Microsoft-built MAI models, and the company’s incentive is to keep all three lines growing rather than to favor one. That is a useful position for a buyer who is worried about lock-in, but it also means Microsoft’s pricing power over any individual lab is partly capped by its own diversification.
For independent labs, the message is sharper. Microsoft is a major commercial partner for frontier labs and is now also their competitor through MAI. Microsoft’s broader catalog reduces its dependence on any single model provider. If one relationship changes, Microsoft can promote MAI or shift workloads to another provider in its catalog.
For self-hosters and open-weight users, the more interesting product is what is getting pulled into Foundry through third-party partnerships. The model catalog Microsoft advertises is broader than just the big three labs, and the agents and middleware Microsoft ships on top of those models are the same packages that any cloud tenant can deploy. Readers who want to stay unaligned can keep running open-weight models on their own hardware; readers who want to compare costs across frontier and in-house models can do that side by side in Foundry, which is a more useful data point than vendor pitch decks.
The OpenAI side is the one quiet risk. Microsoft’s stake is large enough that changes in its valuation show up in Microsoft’s earnings, and the FY2026 full-year gain of $4.96 billion followed a FY2025 loss. The Anthropic Azure spending commitment makes that relationship unusually explicit, while Microsoft’s development of its own frontier models adds another source of competition inside its platform.
The Bottom Line
Microsoft’s Q4 FY2026 earnings put a number on how a hyperscaler can sit on three sides of the AI market at once: $3.2 billion from Anthropic in one quarter, $4.96 billion from OpenAI for the year, and a growing MAI model family of its own. The hedge is working for shareholders, and the pitch to customers is that any model in the stack should be swappable. For readers buying AI through Microsoft, the practical question is no longer which lab to pick, but how much of the bill you want to route through a vendor that is also betting on every lab at once.